Driver risk level identification method based on personality traits and social attributes

Abstract

The invention relates to the technical field of risk assessment, in particular to a driver risk level identification method based on personality characteristics and social attributes, which comprises the following steps: S1, acquiring the personality characteristics and social attributes of a driver; s2, collecting driving behavior data of a driver, and generating an actual driver risk level assessment result; s3, determining key personality characteristics and key social attributes based on a correlation analysis method; s4, integrating key personality characteristics, key social attributes and driver risk level assessment results, and establishing a risk level identification data set; s5, taking the key personality characteristics and the key social attributes as input, taking the risk level of the driver as output, and constructing and training a driver risk level identification model based on a multi-layer perceptron; and S6, verifying the effect of the driver risk level identification model. According to the method, driver risk level identification is carried out by using the easily acquired individual feature data of the driver, the accuracy of risk level identification is ensured, and the operation is simple and convenient and easy to implement.

Description

Technical Field

[0001] The present invention relates to the technical field of risk assessment, and particularly to a method for identifying the risk level of drivers based on personality traits and social attributes. Background Art

[0002] Existing statistical results show that approximately 94% of traffic accidents in the road traffic system are related to human factors. Various dangerous driving behaviors of drivers, such as speeding, rapid acceleration, and rapid deceleration, will significantly increase the risk of accidents. The individual characteristics of drivers, such as driving skill level, psychological characteristics, and emotional fluctuations, are closely related to their driving risks. Therefore, exploring a method for identifying driver risks from the perspective of driver individual characteristics is of great significance for reducing traffic accidents and improving road traffic safety.

[0003] The existing technology mainly uses the method of on-vehicle experiments to identify the risk level of drivers. By collecting the driving behavior performance of drivers in the real road environment, data analysis is carried out on their risk levels. At the same time, real-time physiological state data of drivers during driving, such as heart rate, eye movement, skin conductance, etc., are collected, and a corresponding relationship is established between the real-time physiological state data of drivers and their driving risk levels, and the prediction result of the risk level is obtained through a prediction model. This method has high accuracy, but when collecting data, drivers need to wear physiological information detection devices all the time, which is not easy to operate and may also affect the operation and comfort of drivers.

[0004] Therefore, it is necessary to study a more convenient and easy-to-operate method to predict the risk level of drivers using easily collectible data. Summary of the Invention

[0005] The present invention provides a method for identifying the risk level of drivers based on personality traits and social attributes. This method identifies the risk level of drivers based on easily collectible driver individual characteristic data, not only ensuring the accuracy of the risk level identification result, but also being simple to operate and easy to implement.

[0006] To achieve the above technical effects, the present application provides the following technical solutions:

[0007] A method for identifying the risk level of drivers based on personality traits and social attributes, comprising the following steps:

[0008] S1. Collect the personality traits and social attributes of drivers by using a questionnaire;

[0009] S2. Conduct an on-vehicle driving experiment and collect driving behavior data, and generate an actual driver risk level assessment result based on the dangerous driving behavior data of the driver during driving;

[0010] S3. Determine the key personality traits and key social attributes that have a significant correlation with the driver risk level based on the correlation analysis method;

[0011] S4. Integrate the key personality traits, key social attributes and the actual driver risk level assessment results, establish a risk level identification data set, and divide the risk level identification data set into a training set and a test set;

[0012] S5. Use the key personality traits and key social attributes as inputs and the driver's risk level as the output, construct a driver risk level identification model based on a multi-layer perceptron, and complete the training using the training set;

[0013] S6. Use the test set to verify the identification effect of the driver risk level identification model.

[0014] Technical principle: This application identifies and classifies the driver risk level by constructing a driver risk level identification model, where the input parameters of the risk level identification model use the driver's key personality traits and social attributes as inputs. Collect the driver's multi-dimensional psychological characteristics and social attribute data through psychological scales, and at the same time conduct on-road vehicle experiments to collect the driver's dangerous driving behaviors and evaluate their actual driving risk level data. Determine the driver's key personality traits and social attribute characteristics that have a significant impact on the risk level through the correlation analysis method, and construct a risk level identification model based on MLP, establish the correlation between the driver's personality traits and social attributes and the driving risk level, and output the risk level identification result.

[0015] Beneficial effects: The present invention proposes a driver risk level identification method that is easy to promote and implement. It not only ensures the accuracy of risk level identification, but also takes into account the simplicity of operation and the feasibility of implementation. It can effectively assist the traffic management department to quickly identify high-risk drivers and take corresponding intervention measures (such as safety training, psychological counseling, etc.), thereby effectively preventing and controlling high-risk driving behaviors and reducing the incidence of traffic accidents.

[0016] Further, the S1 includes:

[0017] S11. Use the psychological scale SCL-90 to collect the driver's personality trait data, and the personality trait data includes: somatization, obsessive-compulsive symptoms, interpersonal sensitivity, depression, anxiety, hostility, phobia, paranoia, psychosis;

[0018] S12. Use a questionnaire to collect the driver's social attributes, and the social attributes include: gender, occupational attribute, age, driving experience.

[0019] Beneficial effects: The data provided by professional psychological scales can be used to deeply understand the personality traits of drivers (such as obsessive-compulsive disorder, depression, etc.), which are closely related to driving behavior, so as to better conduct risk assessment and management for different types of drivers. It is simple and easy to operate, and can provide a new method and reference basis for traffic management departments to carry out assessment of drivers' safe driving level and safety education and training.

[0020] Further, the S2 includes:

[0021] S21. Conduct on-road vehicle tests to collect driving behavior data, including speed, acceleration, and heading angle;

[0022] S22. Based on the driving behavior data, use a risk measurement method to quantify the characteristic parameters M(t) of dangerous driving behavior at time t;

[0023] S23. Use the interquartile range method to calculate the threshold of the characteristic parameter of the i-th type of dangerous driving behavior:

[0024]

[0025] where Q i is the upper quartile of the characteristic parameter of the i-th type of dangerous driving behavior, and I i is the interquartile range of the characteristic parameter of the i-th type of dangerous driving behavior;

[0026] S24. According to the proportion of the characteristic parameters of the driver's dangerous driving behavior exceeding the threshold, calculate the driver's dangerous behavior score:

[0027]

[0028] where S i (t) is the instantaneous score of the i-th type of dangerous driving behavior of the driver at time t, and A i is the score of the i-th type of dangerous driving behavior of the driver during the driving period T;

[0029] S25. Normalize the dangerous driving behavior, and the calculation method is as follows:

[0030]

[0031] where min(A i ) is the minimum value among all the dangerous driving behavior scores of the driver, and max(A i ) is the maximum value among all the dangerous driving behavior scores of the driver;

[0032] S26. Perform weighted summation on all the dangerous driving behavior scores of the driver to obtain the comprehensive dangerous driving score of the driver:

[0033]

[0034] In the formula, w i is the weight of the score of the i-th dangerous driving behavior, n is the number of dangerous driving behaviors;

[0035] S27. Based on the comprehensive dangerous driving score of the driver, K-means clustering is adopted, and the sum of squared errors within the cluster and the silhouette coefficient are used as evaluation indicators to obtain the optimal number of classifications of the driver risk level:

[0036]

[0037] In the formula, SSE is the sum of squared errors within the cluster, s is the silhouette coefficient, λ j is the comprehensive dangerous driving score of the j-th sample, F i is the i-th cluster, μ i is the centroid point of the i-th cluster, k is the number of clusters, a(i) is the average distance between sample i and other samples within the same cluster, and b(i) is the average distance between sample i and all samples in other clusters.

[0038] Furthermore, in S26, the CRITIC weight method is used to determine the weights of the dangerous driving behavior scores, and the calculation method is as follows:

[0039]

[0040] In the formula, C i is the amount of information of the i-th dangerous driving behavior, σ i is the standard deviation of the scores of the i-th bad driving behavior of all samples, r ij is the correlation coefficient between the score A Ni of the i-th dangerous driving behavior and the score A Nj of the j-th dangerous driving behavior.

[0041] Furthermore, S3 includes:

[0042] S31. Use the Spearman correlation analysis method to analyze the correlation between the driver's personality traits and their risk level, and extract key personality trait factors;

[0043] S32. Use the non-parametric test method to analyze the relationship between the driver's social attributes and their risk level, and extract key social attributes.

[0044] Beneficial effects: Through correlation analysis and hypothesis testing, it is possible to determine which psychological traits and social attributes have a significant impact on driving risk. This helps to reduce redundant information and improve the identification efficiency and accuracy of the model.

[0045] Further, in S5, the accuracy, precision, recall, and F1 score are used as model performance evaluation indicators to train the optimal identification model, and the index calculation methods are as follows:

[0046]

[0047] In the formula, TP, TN, FP, and FN represent the number of samples of true positive, true negative, false positive, and false negative classes, respectively.

[0048] Further, the key personality traits are obsessive-compulsive symptoms, depression, anxiety, hostility, and paranoia, and the key social attribute is gender.

[0049] Further, the identification model is constructed based on a multi-layer perceptron, including an input layer, a hidden layer, and an output layer, where:

[0050] The input layer contains 6 input nodes for receiving key personality trait and key social attribute data;

[0051] The output layer includes 3 output nodes for outputting the prediction results of the driver risk level;

[0052] The hidden layer includes 10 neuron nodes. Each neuron node is respectively connected to all the input nodes of the input layer and has different weights and biases. The linear combination result generated by the hidden layer is connected to the input end of the non-linear activation function, and the output of the non-linear activation function is connected to the 3 output nodes of the output layer;

[0053] The non-linear activation function uses the Sigmoid function;

[0054] The training set is used to train the identification model. The mean square error is used as the loss function, and the weights and bias coefficients of each neuron in the hidden layer are updated through the backpropagation algorithm to obtain the optimal identification model.

[0055] Beneficial effects: Using a multi-layer perceptron (MLP) as a classifier can extract useful information from a large amount of complex data and achieve high-precision risk level prediction. Compared with traditional linear models, MLP has stronger non-linear fitting ability and can capture more potential patterns. The trained MLP model can quickly complete the processing of a large amount of driver data and risk level identification and prediction, which is suitable for large-scale applications. In addition, with the accumulation of more data, the model can be continuously optimized through incremental learning to further improve efficiency and accuracy. Description of the Drawings

[0056] Figure 1 It is a flow chart of a driver risk level identification method based on personality traits and social attributes;

[0057] Figure 2 It is a structural diagram of a driver risk level identification model based on personality traits and social attributes. Specific implementation manners

[0058] The following is a further detailed description through specific implementation manners:

[0059] Example 1

[0060] This example provides a method for identifying the driver risk level based on personality traits and social attributes. As Figure 1 shown, it includes the following steps:

[0061] S1. Use questionnaires to collect the personality traits and social attributes of drivers;

[0062] Personality traits refer to the relatively stable thinking, emotional, and behavioral patterns of an individual, and these traits can be evaluated through psychological scales. Personality trait data may have an impact on a driver's driving tendency and risky driving behavior. At the same time, the social attribute data of drivers is also closely related to their driving behavior. For example, male drivers may have more aggressive driving behaviors, young drivers may be more prone to speeding or frequent lane-changing, and drivers in high-stress occupations may be more impatient during driving.

[0063] In this example, 50 drivers were recruited to participate in the experiment. Before the experiment started, professional psychological scale SCL-90 was used to comprehensively collect the personality trait data of drivers, and questionnaires were used to collect the social attributes of drivers. Specifically as follows:

[0064] S11. Use the psychological scale SCL-90 to collect the personality trait data of drivers. The personality trait data includes: somatization, obsessive-compulsive symptoms, interpersonal sensitivity, depression, anxiety, hostility, phobia, paranoia, psychosis; among them, somatization reflects symptoms related to physical discomfort, obsessive-compulsive symptoms reflect recurring obsessive thoughts and behaviors, interpersonal sensitivity reflects uncomfortable, inferior, ashamed and other emotions in interpersonal communication, depression reflects depressive symptoms such as low mood and decreased interest, anxiety reflects anxiety symptoms related to tension, worry, uneasiness, etc., hostility reflects emotional reactions such as irritability, anger, and hostility, phobia reflects fear of specific situations or objects, paranoia reflects paranoid thinking such as distrust, suspicion, and suspicion of others, psychosis reflects experiences related to psychotic symptoms such as hallucinations and delusions, and others mainly reflect items related to diet and sleep. The scores of each factor of SCL-90 reflect the severity of such symptoms, and the higher the score, the more obvious such symptoms are.

[0065] The collection results of the psychological scale SCL-90 in this example are shown in Table 1:

[0066] Table 1

[0067]

[0068] S12. Collect the social attributes of drivers by using a questionnaire survey. The social attributes include: gender, occupational attribute, age, and driving experience.

[0069] S2. Conduct on-road driving tests and generate an actual driver risk level assessment result based on the dangerous driving behavior data shown by the drivers during the driving process, specifically as follows:

[0070] S21. Collect driving behavior data during the on-road test, including speed, acceleration, and heading angle.

[0071] In this embodiment, considering that mountain roads, especially two-lane mountain roads, contain many curves and longitudinal slopes and have complex road conditions, more steering, acceleration, braking and other operations are required during driving on such roads, which can enable drivers to actively expose more dangerous driving behaviors. Therefore, this type of road is selected to conduct on-road tests. A variety of in-vehicle sensors are installed on the test vehicle. When the driver is driving on the selected mountain section, the data of the vehicle's speed, acceleration, and heading angle are collected and recorded throughout the process.

[0072] S22. Based on the driving behavior data, use a risk measurement method to quantify the dangerous driving behavior characteristic parameter M(t) at time t.

[0073] In this embodiment, five dangerous driving behaviors, namely speeding, sudden acceleration, sudden deceleration, sudden steering, and unstable speed, are selected, and the corresponding characteristic parameter expressions are as follows:

[0074] Speeding M 1 (t): Select the instantaneous speed of the vehicle as the characteristic parameter, that is, M 1 (t) = v(t);

[0075] Sudden acceleration M 2 (t): Use the positive acceleration of the vehicle as the characteristic parameter to characterize the driver's stepping force on the vehicle throttle, that is, M 2 (t) = a y (t)a y (t)>0;

[0076] Sudden deceleration M 3 (t): Use the negative acceleration of the vehicle as the characteristic parameter to characterize the driver's stepping force on the brake pedal, that is, M 3 (t) = a y (t)a y (t)<0;

[0077] Sudden steering M 4(t): Using the rate of change of the vehicle's heading angle as a characteristic parameter to characterize the urgency of the driver's steering operation, that is where θ(t) is the heading angle of the vehicle at time t;

[0078] Unstable speed M 5 (t): Using the standard deviation of the vehicle's speed within a 3s time window as a characteristic parameter to characterize the degree of the driver's frequent acceleration or deceleration, that is M 5 (t) = std(v(t - 30), v(t - 29),..., v(t)), where the acquisition frequency of speed data is 10Hz, and a total of 30 data points are obtained within the 3s time window.

[0079] S23. Calculate the threshold of the characteristic parameter of the i-th type of dangerous driving behavior using the interquartile range method:

[0080]

[0081] In the formula, Q i is the upper quartile of the characteristic parameter of the i-th type of dangerous driving behavior, and I i is the interquartile range of the characteristic parameter of the i-th type of dangerous driving behavior;

[0082] S24. Calculate the dangerous behavior score of the driver according to the proportion of the characteristic parameter of the driver's dangerous driving behavior exceeding the threshold:

[0083] According to the threshold of the characteristic parameter of the dangerous driving behavior determined by S23, the part exceeding the threshold is defined as the dangerous driving behavior. For example, when M 1 (t) exceeds the threshold, it represents that the driver has committed a speeding behavior. When M 2 (t), M 3 (t) exceed the threshold, it represents that the driver has committed sudden acceleration and sudden deceleration behaviors. When M 4 (t) exceeds the threshold, it represents that the driver has committed a sharp turning behavior. When M 4 (t) exceeds the threshold, it represents that the driver has committed a sharp turning behavior. When M 5 (t) exceeds the threshold, it represents that the driver has committed frequent acceleration and deceleration behaviors within a short period.

[0084] Score all the dangerous driving behaviors committed by the driver during the driving period. The calculation method is as follows:

[0085]

[0086] In the formula, S i (t) is the instantaneous score of the i-th type of dangerous driving behavior of the driver at time t, and A i is the score of the i-th type of dangerous driving behavior of the driver during the driving period T.

[0087] S25. Normalize the i-th type of dangerous driving behavior of all drivers, and the calculation method is as follows:

[0088]

[0089] In the formula, min(A i ) is the minimum value among the scores of all drivers' dangerous driving behaviors, and max(A i ) is the maximum value among the scores of all drivers' dangerous driving behaviors.

[0090] S26. Perform weighted summation on the scores of all drivers' dangerous driving behaviors to obtain the comprehensive dangerous driving score of the driver:

[0091]

[0092] In the formula, w i is the weight of the score of the i-th type of dangerous driving behavior, n is the number of dangerous driving behaviors;

[0093] Among them, the CRITIC weight method is used to determine the weight of the score of the dangerous driving behavior, and the calculation method is as follows:

[0094]

[0095] In the formula, C i is the amount of information of the i-th type of dangerous driving behavior, σ i is the standard deviation of the scores of the i-th type of bad driving behavior of all samples, and r ij is the correlation coefficient between the score A Ni of the i-th type of dangerous driving behavior and the score A Nj of the j-th type of dangerous driving behavior.

[0096] In this embodiment, the weights of the five types of dangerous driving behaviors of the driver are 0.14, 0.20, 0.13, 0.34, and 0.19 in sequence. The normalization results of the scores of the dangerous driving behaviors and the comprehensive dangerous driving scores are shown in Table 2:

[0097] Table 2

[0098]

[0099] S27. Based on the comprehensive dangerous driving score of the driver, obtain the optimal number of classifications of the driver risk level.

[0100] Adopt the K-means clustering analysis method, and use the within-cluster sum of squared errors and the silhouette coefficient as evaluation indicators. The calculation methods are as follows:

[0101]

[0102] Wherein, SSE is the sum of squared errors within clusters, s is the silhouette coefficient, λ j is the comprehensive dangerous driving score of the j-th sample, F i is the i-th cluster, μ i is the centroid point of the i-th cluster, k is the number of clusters, a(i) is the average distance between sample i and other samples within the same cluster, and b(i) is the average distance between sample i and all samples in other clusters.

[0103] In this embodiment, the sum of squared errors and silhouette coefficients corresponding to different numbers of clusters in K-means are shown in Table 3, and the optimal number of classifications is obtained as 3, that is, the driver risk levels are divided into three levels: low risk, medium risk, and high risk.

[0104] Table 3

[0105]

[0106]

[0107] S3. Determine the key personality traits and key social attributes that have a significant correlation with the driver risk level based on the correlation analysis method, including:

[0108] S31. Analyze the correlation between the driver personality traits and their risk levels using the Spearman correlation analysis method, and extract the personality trait factors with significant correlation.

[0109] In this embodiment, the correlation analysis results are shown in Table 4.

[0110] Table 4

[0111]

[0112] As can be seen from Table 4, in the correlation analysis results of personality traits, the P-values of obsessive-compulsive symptoms, depression, anxiety, hostility, and paranoia are all less than 0.05. Therefore, obsessive-compulsive symptoms, depression, anxiety, hostility, and paranoia are selected as the key personality traits.

[0113] S32. Analyze the relationship between the driver social attributes and their risk levels using the non-parametric test method, and extract the social attributes that have a significant impact on the driver risk level.

[0114] In this embodiment, the non-parametric test results are shown in Table 5.

[0115] Table 5

[0116]

[0117] As can be seen from Table 5, the key social attribute that has a significant impact on the driver risk level is gender.

[0118] S4. Integrate the key personality traits, key social attributes, and the evaluation results of the actual driver risk levels, establish a risk level identification data set, and divide the risk level identification data set into a training set and a test set;

[0119] The constructed risk level identification data set includes driver names, driver personality trait data, driver social attribute data, and driver risk level assessment results. The risk level identification data set is divided into a training set and a test set according to 8:2. In this embodiment, the risk level identification data set contains 50 records, of which 40 records are divided into the training set and 10 records are divided into the test set.

[0120] S5. Using the key personality traits and key social attributes as inputs and the driver's risk level as the output, construct a driver risk level identification model based on a multi-layer perceptron and complete the training using the training set;

[0121] The driver risk level identification model is constructed based on a multi-layer perceptron and includes an input layer, a hidden layer, and an output layer. The input layer is used to receive key personality features and key social attribute data. The hidden layer is located between the input layer and the output layer and includes several neuron nodes. Each neuron receives the inputs from all units of the input layer, performs a weighted sum with weights and adds a bias to obtain a linear combination result. The linear combination result is connected to a non-linear activation function as its input, and the classification output result of the non-linear activation function is connected to the neurons of the output layer. The output of the identification model is the predicted result of the driver risk level. The specific structure of the identification model is as Figure 2 shown.

[0122] As described above, the inputs of the identification model include key personality traits and key social attributes, namely, six input variables: obsessive-compulsive symptoms, depression, anxiety, hostility, paranoia, and gender. The driver risk level is divided into three levels: low risk, medium risk, and high risk. Therefore, in this embodiment, the input layer of the identification model contains 6 nodes, the output layer includes 3 nodes, and the hidden layer includes 10 neuron nodes. The non-linear activation function uses the Sigmoid function.

[0123] Use the training set to train the identification model to obtain an optimal identification model. Among them, the loss function of the identification model uses the mean square error function, and the weights and bias coefficients of each neuron in the hidden layer are updated through the backpropagation algorithm.

[0124] S6. Use the test set to verify the identification effect of the driver risk level identification model.

[0125] Using the test set as the input of the optimal identification model, the constructed identification model uses accuracy, precision, recall, and F1 score as the model performance evaluation indicators. The calculation methods of the indicators are as follows:

[0126]

[0127] In the formula, TP, TN, FP, and FN represent the number of samples of true positive, true negative, false positive, and false negative classes respectively.

[0128] The calculated accuracy is 86.7%, the precision is 90.5%, the recall is 86.6%, and the F1 value is 87%, indicating that the model can identify the driver risk level and has a high identification accuracy.

[0129] The above are only the embodiments of the present invention. The present invention is not limited to the fields involved in this embodiment. Common knowledge such as the specific structures and characteristics known in the solutions is not described in detail here. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope required by this application should be based on the content of its claims, and the specific implementation manners described in the specification can be used to interpret the content of the claims.

Claims

1. A driver risk level identification method based on personality traits and social attributes, characterized by: The following steps are involved: S1. Collect drivers’ personality traits and social attributes using questionnaires; S2. Conduct real-vehicle driving tests and collect driving behavior data, and generate actual driver risk level assessment results based on the driver's dangerous driving behavior data during driving; S3. Determine the key personality traits and key social attributes that are significantly correlated with the driver's risk level based on correlation analysis methods; S4. Integrate key personality traits, key social attributes and actual driver risk level assessment results to establish a risk level identification dataset, and divide the risk level identification dataset into a training set and a test set; S5. Using key personality traits and key social attributes as input and the driver's risk level as output, a driver risk level identification model is constructed based on a multi-layer perceptron, and training is completed using a training set; S6. Use the test set to verify the recognition effect of the driver risk level recognition model.

2. A driver risk level identification method based on personality traits and social attributes according to claim 1, characterized in that: The S1 includes: S11. Collecting the driver's personality trait data using the psychology scale SCL-90, wherein the personality trait data include: somatization, obsessive-compulsive symptoms, interpersonal sensitivity, depression, anxiety, hostility, terror, paranoia, and psychosis; S12. Collecting social attributes of drivers using a questionnaire, wherein the social attributes include: gender, occupational attributes, age, and driving experience.

3. A driver risk level identification method based on personality traits and social attributes according to claim 2, characterized in that: The S2 includes: S21. Conduct real vehicle tests and collect driving behavior data, including speed, acceleration, and heading angle; S22, based on the driving behavior data, using a risk measurement method to quantify the dangerous driving behavior characteristic parameter M(t) at time t; S23. Calculate the threshold of the characteristic parameter of the i-th dangerous driving behavior using the interquartile range method: In the formula, Q i is the upper quartile of the characteristic parameters of the i-th dangerous driving behavior, I i is the interquartile range of characteristic parameters of the i-th dangerous driving behavior; S24. Calculate the driver's dangerous behavior score based on the proportion of the driver's dangerous driving behavior characteristic parameters exceeding the threshold: In the formula, S i (t) is the instantaneous score of the driver’s i-th dangerous driving behavior at time t, A i is the score of the driver’s i-th dangerous driving behavior during driving period T; S25. Normalize the dangerous driving behavior, and the calculation method is as follows: Where, min(A i ) is the minimum value of all dangerous driving behavior scores of the driver, max(A i ) is the maximum value of all the dangerous driving behavior scores of the driver; S26. Perform a weighted summation of all the dangerous driving behavior scores of the driver to obtain the driver's comprehensive dangerous driving score: In the formula, w i is the weight of the score of the i-th dangerous driving behavior, n is the number of dangerous driving behaviors; S27. Based on the comprehensive dangerous driving score of the driver, K-means clustering is used, and the sum of squared errors and silhouette coefficient within the cluster are used as evaluation indicators to obtain the optimal classification number of the driver's risk level: In the formula, SSE is the sum of squared errors within the cluster, s is the silhouette coefficient, and λ is j is the comprehensive dangerous driving score of the jth sample, F i is the i-th cluster, μ i is the centroid of the ith cluster, k is the number of clusters, a(i) is the average distance between sample i and other samples in the same cluster, and b(i) is the average distance between sample i and all samples in other clusters.

4. A driver risk level identification method based on personality traits and social attributes according to claim 3, characterized in that: In S26, the CRITIC weight method is used to determine the dangerous driving behavior score weight, and the calculation method is as follows: In the formula, C i is the information amount of the i-th dangerous driving behavior, σ i is the standard deviation of the i-th bad driving behavior scores of all samples, r ij Score A for the i-th dangerous driving behavior Ni and the j-th dangerous driving behavior score A Nj The correlation coefficient between .

5. The method for identifying driver risk level based on personality traits and social attributes according to claim 4, characterized in that: The S3 includes: S31. Use the Spearman correlation analysis method to analyze the correlation between the driver's personality traits and their risk level, and extract key personality trait factors; S32. Use non-parametric test methods to analyze the relationship between drivers’ social attributes and their risk levels and extract key social attributes.

6. The method for identifying driver risk level based on personality traits and social attributes according to claim 5, characterized in that: The S5 uses accuracy, precision, recall and F1 score as model performance evaluation indicators to train the optimal recognition model. The indicator calculation method is as follows: Where TP, TN, FP and FN represent the number of samples of true positive class, true negative class, false positive class and false negative class respectively.

7. The method for identifying driver risk level based on personality traits and social attributes according to claim 1, characterized in that: The key personality traits are obsessive-compulsive symptoms, depression, anxiety, hostility, and paranoia, and the key social attribute is gender.

8. The method for identifying driver risk level based on personality traits and social attributes according to claim 1, characterized in that: The driver risk level is constructed based on a multi-layer perceptron, including an input layer, a hidden layer and an output layer, where: The input layer contains 6 input nodes, which are used to receive key personality traits and key social attribute data; The output layer includes 3 output nodes, which are used to output the prediction results of the driver’s risk level; The hidden layer includes 10 neuron nodes, each of which is connected to all input nodes of the input layer and has different weights and biases. The linear combination result generated by the hidden layer is connected to the input end of the nonlinear activation function, and the output of the nonlinear activation function is connected to the three output nodes of the output layer. The nonlinear activation function uses the Sigmoid function; The recognition model is trained using the training set, the mean square error is used as the loss function, and the weights and biases of the hidden layer neurons are updated through the back propagation algorithm to obtain the optimal recognition model.

Images