Driving style recognition method for driving risk assessment, electronic device, medium

Abstract

This invention discloses a driving style identification method, electronic device, and medium for driving risk assessment, comprising: acquiring vehicle driving operation data, vehicle status data, and accident records for a certain car insurance period; using driving operation data as samples and accident records as sample labels, preprocessing and standardizing the samples based on vehicle status data; inputting driving operation data and corresponding speed sequences into a pre-trained driving style feature extraction model to extract driving features, including driving style features and speed mapping; clustering the extracted driving features; obtaining the silhouette coefficient, accident probability standard deviation, and sample distribution of each clustering result, obtaining a clustering result score for each clustering result based on the silhouette coefficient, accident probability standard deviation, and sample distribution, and selecting the clustering result with the highest clustering result score as the optimal clustering result; and classifying the accident risk level according to the accident probability based on the optimal clustering result.

Description

Technical Field

[0001] This invention belongs to the field of driving behavior data processing, and particularly relates to a driving style identification method, electronic device, and medium for driving risk assessment. Background Technology

[0002] Driving style refers to a driver's unique and habitual way of operating a vehicle, reflecting their driving tendencies, preferences, and personalized driving patterns. Although differences in individual attributes and the traffic and road environments they face lead to significant variations in driving operations, some commonalities exist that allow us to categorize drivers into different driving styles. In-depth research into driving styles and the development of accurate driving style identification models are of great significance for improving traffic safety and designing personalized vehicle services.

[0003] Existing research typically utilizes prior knowledge to design experiments or employs expert evaluation methods to obtain sample sets labeled with driving styles, making it difficult to guarantee the objectivity of the sample sets. Furthermore, existing research often overlooks the impact of traffic and road conditions on driver operation, resulting in extracted driving style features that do not accurately reflect the driver's true style. Finally, current research tends to categorize driving styles into cautious, moderate, and aggressive types; however, these classification methods are often not directly applicable to specific real-world scenarios, especially in the vehicle insurance industry. Summary of the Invention

[0004] In view of this, the present invention proposes a driving style identification method, electronic device, and medium for driving risk assessment.

[0005] In a first aspect, embodiments of the present invention provide a driving style identification method for driving risk assessment, the method comprising:

[0006] To obtain vehicle driving operation data, vehicle status data, and accident records for a specific car insurance period;

[0007] Driving operation data is used as a sample, and accident records are used as sample labels. The samples are preprocessed and standardized based on vehicle status data.

[0008] Driving operation data and corresponding speed sequences are input into a pre-trained driving style feature extraction model to extract driving features, which include driving style features and speed mappings. The loss function corresponding to the driving style feature extraction model is the sum of the mean square error between the input driving operation data and the driving operation data predicted by the driving style feature extraction model, the mean square error between the input speed sequence and the speed sequence predicted by the driving style feature extraction model, and the mean square error between the speed sequence and speed mapping specified in the World Light Vehicle Test Cycle.

[0009] Cluster the extracted driving features;

[0010] The process involves obtaining the silhouette coefficient, standard deviation of accident probability, and sample distribution for each clustering result. Based on these parameters, a clustering result score is assigned to each clustering result, and the clustering result with the highest score is selected as the optimal clustering result. The calculation of the standard deviation of accident probability includes: all samples are clustered into several driving style categories; based on the accident records of each sample in the current car insurance period, the average accident probability of each driving style category is calculated, and the standard deviation of the accident probability is obtained.

[0011] The risk level is classified according to the probability of occurrence based on the optimal clustering results.

[0012] Secondly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the above-described driving style identification method for driving risk assessment.

[0013] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the above-described driving style identification method for driving risk assessment.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0015] 1. This invention collects a batch of multi-dimensional driving operation data (steering wheel angle, gear position, accelerator pedal travel and brake pedal travel) from the vehicle's CAN-BUS during actual vehicle use to form a raw dataset to reflect the driving styles of different drivers.

[0016] 2. A convolutional neural network with self-attention mechanism is used to perform feature dimensionality reduction on multidimensional driving operation data, which can extract comprehensive spatial and temporal features.

[0017] 3. By adding a standardization function to the autoencoder, different drivers can be constrained to the same driving speed sequence, realizing the extraction of driving style characteristics of different drivers under the same working conditions. Using the mean square error between the speed sequence and speed mapping specified by the World Light Vehicle Test Cycle as the constraint condition, the WLTC test cycle is used to restore common domestic working conditions, so that the obtained driving style characteristics can characterize driving risks.

[0018] 4. This invention selects an appropriate number of clusters by comprehensively considering the profile coefficient, the standard deviation of the accident probability, and the sample distribution. It also incorporates actual vehicle insurance accident data from the driver's previous insurance year to calculate the accident probability for drivers with different driving styles. The results show that the feature extraction algorithm developed in this paper can accurately extract features representing driving risk, and drivers with different driving styles exhibit significant differences in accident probability. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of a driving style identification method for driving risk assessment provided in an embodiment of the present invention;

[0021] Figure 2 This is a structural diagram of the self-encoder provided in an embodiment of the present invention;

[0022] Figure 3 A schematic diagram of the autoencoder training results provided in an embodiment of the present invention;

[0023] Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] It should be noted that, unless otherwise specified, the features in the following embodiments and implementation methods can be combined with each other.

[0026] This invention provides a driving style identification method for driving risk assessment, such as... Figure 1 As shown, the method specifically includes the following steps:

[0027] Step S1: Obtain the vehicle's driving operation data, vehicle status data, and accident records for a specific car insurance period.

[0028] Specifically, the driving operation data refers to the data generated by the driver operating the vehicle's steering wheel, gear shift, accelerator pedal, and brake pedal, including steering wheel angle, gear position, accelerator pedal travel value, and brake pedal travel value; the vehicle status data includes vehicle speed, timestamp, and cumulative mileage. The accident record indicates whether the driver had any accidents in the previous insurance period; if there was an accident, it is marked as 1, otherwise it is marked as 0.

[0029] Furthermore, in this example, the vehicle's driving operation data and vehicle status data during a certain insurance period are collected through the passenger vehicle controller area network (CAN) bus and stored in a cloud database.

[0030] It should be noted that the Control Area Network (CAN) bus in passenger vehicles collects multi-dimensional data generated when the driver operates in-vehicle devices (steering wheel, gear shift, accelerator pedal, brake pedal), such as steering wheel angle, gear position, accelerator pedal travel, and brake pedal travel. Driving operation data is directly generated by the driver's actions and can most directly reflect the driver's driving style. Drivers typically operate multiple in-vehicle devices simultaneously depending on driving conditions; therefore, these multi-dimensional driving operation data (steering wheel angle, gear position, accelerator pedal travel, brake pedal travel) are not isolated from each other. Comprehensive analysis of this data can provide a more complete characterization of the driver's driving style.

[0031] Step S2: Use driving operation data as samples and accident records as sample labels, and preprocess and standardize the samples based on vehicle status data.

[0032] Specifically, step S2 includes the following sub-steps:

[0033] Step S201: Slice the continuous data in the sample into segments according to a preset length, calculate the derived value of each segment, and filter and delete abnormal data and abnormal derived values.

[0034] In this example, the continuous data in the sample refers to data with a sampling time interval of 10 seconds, the slice length is set to 3600, and each data segment contains 3600 data points.

[0035] The derived values ​​include cumulative mileage, vehicle speed standard deviation, cumulative duration, and the ratio of cumulative mileage to cumulative duration.

[0036] Furthermore, filtering and deleting abnormal data and abnormal derived values ​​includes: deleting data segments with a speed standard deviation of 0, as this indicates that the vehicle speed did not change throughout the time period and should be considered erroneous data; and deleting data segments with a cumulative mileage to cumulative time ratio greater than 220, as this indicates that there may be jumps in the statistics of the vehicle's cumulative mileage.

[0037] Step S202: The Z-Score algorithm is used to standardize the data preprocessed in step S201. The expression is as follows:

[0038]

[0039] In the formula, Represents the original data. For raw data The mean, For raw data standard deviation This is the standardized result.

[0040] Step S3: Input the driving operation data and the corresponding speed sequence into the pre-trained driving style feature extraction model. The driving style feature extraction model extracts driving features, which include driving style features and speed mapping. The loss function corresponding to the driving style feature extraction model is the sum of the mean square error between the input driving operation data and the driving operation data predicted by the driving style feature extraction model, the mean square error between the input speed sequence and the speed sequence predicted by the driving style feature extraction model, and the mean square error between the speed sequence and speed mapping specified in the World Light Vehicle Test Cycle.

[0041] like Figure 2 As shown, in this example, the driving style feature extraction model uses a convolutional autoencoder, which includes an encoder part and a decoder part. The encoder part includes an input layer, a first convolutional layer, a first maxpooling layer, a second convolutional layer, a second maxpooling layer, a self-attention mechanism module, and two fully-connected layers (FC layers) connected in sequence. The decoder part includes two fully connected layers, a first unmaxpooling layer, a first unconvolutional layer, a second unmaxpooling layer, a second deconvolutional layer, and an output layer connected in sequence.

[0042] The self-attention mechanism module consists of a self-attention layer and a learnable parameter. Assume the input to the self-attention layer is... The output of the self-attention layer is The learnable parameters are The output of the self-attention mechanism is The self-attention mechanism can be represented as:

[0043]

[0044] Furthermore, the driving features extracted by the driving style feature extraction model can be decomposed into driving style features f and speed mapping v. The relationship between the three can be expressed as:

[0045]

[0046]

[0047]

[0048] Furthermore, the loss function corresponding to the driving style feature extraction model is: input driving operation data Driving operation data predicted by the driving style feature extraction model The mean square error between them, the input velocity sequence Speed ​​sequence predicted by driving style feature extraction model The mean square error between them, and the speed sequence specified in the world light vehicle test cycle. The sum of the mean squared errors between the velocity mapping v and the mean squared error. The expression is as follows:

[0049]

[0050] In the formula, For the sample size, , This test cycle uses the speed sequence specified in the Worldwide Harmonized Light-duty Vehicles Test Cycle (WLTC). This test cycle can characterize the general features of driving conditions in China. Using the mean squared error between the speed sequence specified in the WLTC and the speed mapping as a constraint, the convergence of the loss function during training ensures that each driver can achieve the driving sequence specified in the WLTC, enabling the extraction of features from different drivers under the same driving conditions.

[0051] Furthermore, the training results of the driving style feature extraction model are as follows: Figure 3 As shown, the driving style feature extraction model can perform speed mapping very well. The consistency with the WLTC velocity sequence is such that the error of more than 95% of the sample points is between (-2.0km / h, 2.0km / h), which can meet the application needs of various industries.

[0052] It should be noted that the driving style features extracted by the autoencoder are time-series data of length 1800. The use of convolutional layers can improve the model's ability to extract features from neighboring regions of multi-dimensional data, including spatial features between data of different dimensions and temporal features between data at different time steps. During driving, drivers may encounter similar traffic and road environments at varying intervals, leading to similar driving actions. The use of self-attention mechanisms can improve the model's ability to extract features from multi-dimensional data at indefinite distances, enabling better extraction of such features.

[0053] Step S4: Cluster the driving features extracted in step S3.

[0054] Specifically, in this example, the TimeSeriesKMeans algorithm is used to cluster the driving features extracted in step S3. TimeSeriesKMeans clustering is a machine learning method that iteratively seeks the optimal clustering result step by step based on data iteration. It includes the following steps:

[0055] Step S401: First, specify the number of clusters k, specify the number of iterations, and randomly select k cluster centers.

[0056] Step S402: Next, the distance between each sample and the cluster center is calculated according to the softDTW algorithm, and the samples with the closest distance are grouped into one class.

[0057] Step S403: Then determine whether the error between the two iterations is within the preset average error range. If it is not, update the cluster centers, redistribute the samples, and calculate the average error. If it is, end the iteration.

[0058] Step S404: The final output is the clustering result after the iteration ends or is completed.

[0059] Step S5: Obtain the silhouette coefficient, standard deviation of accident probability, and sample distribution for each clustering result. Based on the silhouette coefficient, standard deviation of accident probability, and sample distribution, obtain the clustering result score for each clustering result. The clustering result with the highest clustering result score is taken as the optimal clustering result. The calculation process of the standard deviation of accident probability includes: all samples are clustered into several driving style categories. Based on the accident records of each sample in the car insurance cycle, calculate the average accident probability of each driving style category to obtain the standard deviation of accident probability.

[0060] Specifically, the standard deviation of the probability of occurrence (SC) includes:

[0061] The standard deviation of the accident probability is used to characterize the difference between the average accident probabilities of each category in the current clustering results. The larger the standard deviation of the accident probability SC, the greater the difference in the accident probability of different driving styles, and the more accurate the style identification.

[0062] After clustering, all samples Clustered into , ..., There are k driving style categories. This indicates that the sample had no claims record in the previous insurance period. This indicates that the sample had an accident record in the previous insurance period, and the average accident probability for each driving style category. It can be represented as:

[0063]

[0064] In the formula, This represents the total number of samples belonging to driving style category k.

[0065] The standard deviation of the probability of occurrence can be expressed as:

[0066]

[0067] In the formula, This represents the average probability of an accident across all categories.

[0068] The sample distribution SD includes:

[0069] The sample distribution is used to characterize the difference in the distribution of the number of samples in each driving style category in this clustering result. Generally, the smaller the sample distribution SD, the more uniform the distribution of samples in each driving style category.

[0070] The sample distribution SD can be represented as:

[0071]

[0072] In the formula, This represents the average number of samples across all driving style categories.

[0073] The cluster score σ for each clustering result is obtained based on the silhouette coefficient, the standard deviation of the probability of occurrence, and the sample distribution. The expression is as follows:

[0074]

[0075] In the formula, SI represents the profile coefficient. As a weighting factor ( The value is the number of driving style categories, where... ,SC, All samples have undergone sum-normalization standardization. The specific implementation method (taking the sample distribution SD as an example) is as follows:

[0076]

[0077] In the formula, j represents the number of clustering operations performed to compare the clustering results.

[0078] It should be noted that while drivers exhibit similar driving styles, their individual characteristics still exist. Furthermore, the effectiveness of clustering results is guaranteed once the silhouette coefficient exceeds a certain range. Therefore, the square root of the silhouette coefficient SI is used to calculate the clustering result score σ to reduce the impact of excessively high silhouette coefficients. Since increasing the number of clusters inevitably leads to a decrease in the sample distribution SD, a weighting factor is added... The method of constraining the number of clusters is adopted. Calculate the clustering result score σ. The larger the clustering result score σ, the better the clustering result.

[0079] For example, this embodiment collected and analyzed relevant data from 73 drivers in the previous insurance year. After feature extraction and clustering, the results for clusters with 3, 4, and 5 clusters are shown in Tables 1 and 2, respectively. The accident probability is the average accident probability of all samples in that category.

[0080] Table 1

[0081]

[0082] Table 2

[0083]

[0084] Table 3

[0085]

[0086] Calculations show that when k=3, σ=0.336; when k=4, σ=0.341; and when k=5, σ=0.323. The clustering result score σ is highest when k=4; therefore, the clustering result with 4 clusters is selected as the final result.

[0087] It should be noted that traditional unsupervised clustering often determines the optimal number of clusters solely based on the silhouette coefficients of different cluster sizes, easily overlooking the actual characteristics of the sample data. While drivers may exhibit similar driving styles, their individual characteristics still exist. Therefore, this invention considers not only the silhouette coefficients but also the standard deviation of the accident probability and the sample distribution when selecting the optimal number of clusters. Because this invention extracts the characteristics of different drivers under the same traffic and road conditions, the accident probabilities of drivers with different driving styles should be significantly different, reflected in the magnitude of the standard deviation of the accident probability. Since this embodiment uses entirely real-world driver data, some extreme samples may appear; therefore, the sample distribution in the clustering results also needs to be considered.

[0088] As shown in Tables 1 and 2, when comparing the clustering results with 3 and 4 clusters respectively, the silhouette coefficient of the former is significantly larger than that of the latter. However, both silhouette coefficients are at a high level, proving that both clustering results are effective. Furthermore, the standard deviations of the accident probabilities are similar, indicating that both clustering methods can distinguish driving styles with significantly different accident probabilities. Considering the sample distribution in the two clustering results, it can be seen that both clustering methods can clearly identify two extreme groups of samples. When the number of clusters is only 3, the TimeSeriesKMeans algorithm is insufficient to distinguish most samples. However, when the number of clusters is 4, apart from the two extreme groups, the remaining samples are reasonably divided into two different driving styles, and the accident probabilities of the different driving styles are significantly different, at 0.267 and 0.473 respectively.

[0089] As shown in Table 3, when the number of clusters is 5, the clustering results are not better than those when the number of clusters is 4, in terms of silhouette coefficient, standard deviation of probability of occurrence, and sample distribution. Therefore, considering all factors, this embodiment should select the clustering results when the number of clusters is 4 as the final result.

[0090] Of the three indicators proposed in this invention, the silhouette coefficient reflects the effectiveness of the clustering results, the standard deviation of the accident probability can verify the accuracy of the clustering results to a certain extent, and the sample distribution can reflect whether there are extreme samples. By comprehensively considering the three indicators and selecting an appropriate number of clusters based on the clustering result scores, the optimal driving style identification result can be obtained.

[0091] Step S6: Based on the optimal clustering results, classify the risk level according to the probability of occurrence.

[0092] Based on the optimal clustering results and the probability of an accident, the risk levels of the accident are marked as: Risk Level 1, Risk Level 2, ..., Risk Level k (k is the number of clusters, i.e. the total number of driving style categories).

[0093] For example, in this case, all samples are divided into four driving styles, with significant differences in the probability of accidents among drivers of different styles. When designing vehicle insurance, drivers with Level 4 and Level 3 driving styles, due to their highest probability of accidents, may be considered for higher premiums, while drivers with Level 2 and Level 1 driving styles, due to their lower probability of accidents, may be considered for lower premiums.

[0094] like Figure 4 As shown, this application provides an electronic device including a memory 101 for storing one or more programs and a processor 102. When the one or more programs are executed by the processor 102, they implement the method as described in any of the first aspects above.

[0095] The system also includes a communication interface 103. The memory 101, processor 102, and communication interface 103 are electrically connected directly or indirectly to each other to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules, and the processor 102 executes various functional applications and data processing by executing the software programs and modules stored in the memory 101. The communication interface 103 can be used for signaling or data communication with other node devices.

[0096] The memory 101 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0097] The processor 102 can be an integrated circuit chip with signal processing capabilities. The processor 102 can be a general-purpose processor 102, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0098] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can also be implemented in other ways. The method and system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0099] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0100] On the other hand, embodiments of this application provide a computer-readable storage medium storing a computer program thereon. When executed by processor 102, the computer program implements the methods described in any of the first aspects above. If the functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0101] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A driving style recognition method for driving risk assessment, characterized by, The method comprises: obtaining driving operation data, vehicle state data and accident record of a vehicle in a certain vehicle insurance period; taking the driving operation data as samples and the accident record as sample labels, and pre-processing and standardizing the samples according to the vehicle state data; inputting the driving operation data and the corresponding speed sequence into a pre-trained driving style feature extraction model to extract driving features, wherein the driving features include driving style features and speed mapping; wherein the loss function corresponding to the driving style feature extraction model is the sum of the mean square error between the input driving operation data and the driving operation data predicted by the driving style feature extraction model, the mean square error between the input speed sequence and the speed sequence predicted by the driving style feature extraction model, and the mean square error between the speed sequence specified in the world light-duty vehicle test cycle and the speed mapping; clustering the extracted driving features; obtaining the silhouette coefficient, the accident probability standard deviation and the sample distribution of each clustering result, obtaining the clustering result score of each clustering result according to the silhouette coefficient, the accident probability standard deviation and the sample distribution, and taking the clustering result with the highest clustering result score as the optimal clustering result; wherein the calculation process of the accident probability standard deviation comprises: all samples are clustered into several driving style categories, the average accident probability of each driving style category is calculated according to the accident record of each sample in the vehicle insurance period, and the accident probability standard deviation is obtained; dividing the accident risk level according to the optimal clustering result according to the accident probability. 2.The driving style recognition method for driving risk assessment according to claim 1, wherein, The driving operation data includes steering wheel angle, gear, accelerator pedal stroke value and brake pedal stroke value; the vehicle state data includes vehicle speed, time stamp and vehicle cumulative mileage; the accident record indicates whether the driver has an accident in the last insurance period, and is marked as 1 if there is an accident, otherwise is marked as 0. 3.The driving style recognition method for driving risk assessment according to claim 1, wherein, The pre-processing of the samples according to the vehicle state data comprises: slicing the continuous data in the samples according to a preset length, calculating the derived value of each data segment, and screening and deleting abnormal data and abnormal derived values; wherein the derived value includes cumulative mileage, vehicle speed standard deviation, cumulative time length, and the ratio of cumulative mileage to cumulative time length; screening and deleting abnormal data and abnormal derived values, including: deleting data segments with vehicle speed standard deviation equal to 0; deleting data segments with cumulative mileage to cumulative time length ratio greater than 220.

4. The driving style recognition method for driving risk assessment according to claim 1, wherein, The driving style feature extraction model adopts a convolutional autoencoder, which includes an encoder part and a decoder part; the encoder part includes an input layer, a first convolutional layer, a first max-pooling layer, a second convolutional layer, a second max-pooling layer, a self-attention mechanism module and two layers of fully connected layers connected in sequence; the decoder part includes two layers of fully connected layers, a first reverse max-pooling layer, a first reverse convolutional layer, a second reverse max-pooling layer, a second reverse convolutional layer and an output layer connected in sequence.

5. The driving style recognition method for driving risk assessment according to claim 1, wherein, The expression of the loss function corresponding to the driving style feature extraction model is: ； wherein is the number of samples, , represents the input driving operation data, represents the driving operation data predicted output by the driving style feature extraction model, represents the input speed sequence, represents the speed sequence predicted output by the driving style feature extraction model, represents the speed sequence prescribed by the World Light-duty Vehicle Test Cycle, v represents the speed mapping.

6. The driving style recognition method for driving risk assessment according to claim 1, wherein, The calculation process of the accident probability standard deviation comprises: After clustering, all samples are divided into , …, k driving style categories, indicates that the sample has no record of accidents in the last insurance period, indicates that the sample has a record of accidents in the last insurance period, and the average accident probability of each driving style category is expressed as: ； wherein denotes the total number of samples belonging to the driving style class k; Standard deviation of the probability of loss The expression for the standard deviation of the probability of loss is: ； In the formula, represents the average probability of loss of all classes.

7. The driving style recognition method for driving risk evaluation according to claim 1 or 6, characterized in that, The expression of the clustering result score of each clustering result obtained according to the silhouette coefficient, the accident probability standard deviation and the sample distribution is as follows: ； In the formula, SI represents a profile coefficient, represents a clustering result score, represents a standard deviation of a risk probability, represents a sample distribution, represents a weight factor, The value of the number of driving style categories; Wherein, the expression of sample distribution SD is as follows: ； wherein denotes the average number of samples of all driving style classes, denotes the total number of samples belonging to driving style class i. 8.The driving style recognition method for driving risk assessment according to claim 1, wherein, According to the optimal clustering result, the risk levels are divided according to the probability of occurrence, including: According to the optimal clustering result, the risk levels are marked according to the probability of occurrence, including: risk level 1, risk level 2, …, risk level k, k is the total number of driving style categories.

9. An electronic device comprising a memory and a processor, characterized in that The memory is coupled with the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to realize the driving style recognition method for driving risk evaluation according to any one of claims 1-8.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to realize the driving style recognition method for driving risk evaluation according to any one of claims 1-8.

Images