Risk control information processing method and system based on Internet of Vehicles

Abstract

The present invention provides a risk control information processing method and system based on the Internet of Vehicles. By comprehensively considering the dynamic driving state characteristics and risk association characteristics of the target Internet of Vehicles user, accurate risk control description knowledge characteristics are generated, thereby achieving a comprehensive and detailed characterization of the user's driving risk. Utilizing the constructed target risk control decision network, the method can generate accurate risk prediction data for multiple risk dimensions, wherein the design of the risk control decision unit and the dynamic adjustment mechanism of the number of network sub-units ensure the accuracy and efficiency of risk prediction. Furthermore, by combining the user's risk control management needs and the prior risk control portrait feature data of the reference risk control response strategy, the present invention can accurately calculate the matching degree of each strategy with the target risk dimension, and tailor the most suitable risk control response strategy for the user. This method not only improves the risk control accuracy and response speed in the Internet of Vehicles environment, but also significantly enhances the pertinence and effectiveness of the risk control strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Vehicles, and in particular to a risk control information processing method and system based on Internet of Vehicles. Background Art

[0002] With the rapid development of Internet of Vehicles technology, the information exchange between vehicles, vehicles and infrastructure, and vehicles and the cloud is becoming more and more frequent, providing a rich source of data and broad application prospects for risk control during vehicle driving. However, how to effectively use these massive data to accurately assess and predict the driving behavior of Internet of Vehicles users, and then formulate a reasonable risk control response strategy, has become a key issue that needs to be solved in the current field of Internet of Vehicles technology.

[0003] Traditional risk control information processing methods often rely on single driving behavior characteristics or simple statistical models, which are difficult to fully reflect the complex driving status and potential risks of Internet of Vehicles users. At the same time, these methods often have limitations such as low prediction accuracy and slow response speed when dealing with multi-dimensional and highly complex risk predictions, making it difficult to meet the high requirements for risk control in the Internet of Vehicles environment. Summary of the invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a risk control information processing method based on the Internet of Vehicles, the method comprising:

[0005] Generate corresponding risk control description knowledge features based on the dynamic driving state features and risk association features of the target Internet of Vehicles user data; wherein the dynamic driving state features reflect the vehicle driving road condition state corresponding to the target Internet of Vehicles user data, and the risk association features reflect the multi-dimensional driving behavior features of the target Internet of Vehicles user data;

[0006] Through the multiple risk control decision units included in the target risk control decision network, based on the risk control description knowledge characteristics, risk prediction data of the target Internet of Vehicles user data in multiple risk dimensions are generated; wherein, one risk control decision unit corresponds to a risk prediction request of one risk dimension, one risk control decision unit generates risk prediction data of one risk dimension, and the number of network subunits of each risk control decision unit is positively correlated with the prediction difficulty of the corresponding risk prediction request, and the target risk control decision network is generated by joint network parameter learning based on the training cost layers corresponding to the multiple risk control decision units respectively;

[0007] Obtaining risk control management requirements for the target Internet of Vehicles user data, and prior risk control profile feature data of multiple reference risk control response strategies in the multiple risk dimensions;

[0008] Generate the risk control response matching degree of the plurality of reference risk control response strategies for the target Internet of Vehicles user data in the target risk dimension according to the risk prediction data of the target risk dimension corresponding to the risk control management requirement and the obtained priori risk control portrait feature data;

[0009] At least one target risk control response strategy for the target Internet of Vehicles user data is determined based on the risk control response matching degree.

[0010] On the other hand, an embodiment of the present invention also provides a risk control information processing system based on the Internet of Vehicles, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0011] Based on the above aspects, the embodiment of the present application generates accurate risk control description knowledge features by comprehensively considering the dynamic driving state characteristics and risk association characteristics of the target Internet of Vehicles user, thereby achieving a comprehensive and detailed characterization of the user's driving risk. Utilizing the constructed target risk control decision network, the method can generate accurate risk prediction data for multiple risk dimensions, wherein the design of the risk control decision unit and the dynamic adjustment mechanism of the number of network sub-units ensure the accuracy and efficiency of risk prediction. Furthermore, by combining the user's risk control management needs and the prior risk control portrait feature data of the reference risk control response strategy, the present invention can accurately calculate the matching degree of each strategy with the target risk dimension, and tailor the most suitable risk control response strategy for the user. This method not only improves the risk control accuracy and response speed in the Internet of Vehicles environment, but also significantly enhances the pertinence and effectiveness of the risk control strategy, providing a safer and more reliable driving environment for Internet of Vehicles users. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of the execution flow of the risk control information processing method based on the Internet of Vehicles provided by an embodiment of the present invention.

[0013] Figure 2 It is a schematic diagram of the hardware architecture of a risk control information processing system based on the Internet of Vehicles provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 It is a flow chart of a risk control information processing method based on the Internet of Vehicles provided by an embodiment of the present invention. The risk control information processing method based on the Internet of Vehicles is introduced in detail below.

[0015] Step S110, generating corresponding risk control description knowledge features based on the dynamic driving state features and risk association features of the target IoV user data. The dynamic driving state features reflect the vehicle driving road condition state corresponding to the target IoV user data, and the risk association features reflect the multi-dimensional driving behavior features of the target IoV user data.

[0016] In this embodiment, for a connected vehicle user named Mr. Zhang, his vehicle is equipped with various sensors, which will continuously collect vehicle-related data and upload it to the connected vehicle system to form target connected vehicle user data.

[0017] For the dynamic driving state characteristics, the vehicle's speed sensor shows that Mr. Zhang's average speed is 30 kilometers per hour when driving on urban roads, but the speed suddenly drops to less than 10 kilometers per hour in some sections, which may be due to encountering traffic lights or congested sections. The data from the acceleration sensor shows that the vehicle has frequent acceleration changes when starting and braking, for example, the acceleration increases rapidly from 0 to 10 kilometers per hour in a short period of time, and then quickly drops to 0. The data from the driving direction sensor shows that the vehicle has multiple turning operations during driving, especially when entering and leaving parking lots or turning intersections. At the same time, the geographic location information obtained by the vehicle's positioning system shows that Mr. Zhang often drives in commercial areas with a high traffic congestion index, and the types of roads he drives on include main roads, branch roads, and some small alleys. The proportion of driving time on main roads is about 60%, and the time spent in congested areas accounts for 30% of the total driving time. These data directly or indirectly reflect the driving road conditions of the vehicle and together constitute the dynamic driving state characteristics.

[0018] From the perspective of risk association characteristics, Mr. Zhang's driving behavior also has multi-dimensional characteristics. Through the analysis of vehicle operation data, it is found that the force distribution characteristics of Mr. Zhang's brake and accelerator operation are that the proportion of light braking and accelerator is relatively high, about 70%, and the proportion of heavy braking is 30%. In terms of the continuity of brake and accelerator operation, the duration of his continuous light braking or accelerator is relatively short, generally not exceeding 3 seconds. From the analysis of driving time regularity characteristics, Mr. Zhang mainly drives in the two time periods of 8 to 9 in the morning and 5 to 6 in the afternoon. The driving time in these two time periods accounts for about 60%, and the standard deviation of driving time is small, indicating that the driving time is relatively regular. From the perspective of driving fatigue characteristics, Mr. Zhang's continuous driving time of more than 2 hours (set time threshold) is rare, only 2 times in the past month, and the frequency of operating errors during driving is also low, about 0.5 times per 100 kilometers. These multi-dimensional driving behavior characteristics constitute risk association characteristics.

[0019] By combining these dynamic driving state characteristics and risk association characteristics, risk control description knowledge characteristics are generated. For example, if the time spent in a congested area is too long and the driving time is concentrated in peak hours, and the frequency of operating errors during driving is low but the proportion of light braking and accelerator is high, the risk control description knowledge characteristics that may be generated are: there is a certain risk during peak congestion periods, but because there are few operating errors and relatively stable operating habits, the risk is controllable but needs attention.

[0020] Step S120, through the multiple risk control decision units included in the target risk control decision network, based on the risk control description knowledge characteristics, generate the risk prediction data of the target Internet of Vehicles user data in multiple risk dimensions. Among them, one risk control decision unit corresponds to a risk prediction request of a risk dimension, one risk control decision unit generates risk prediction data of a risk dimension, and the number of network subunits of each risk control decision unit is positively correlated with the prediction difficulty of the corresponding risk prediction request. The target risk control decision network is generated by joint network parameter learning based on the training cost layers corresponding to the multiple risk control decision units.

[0021] Assume that there are three risk control decision units in the target risk control decision network, corresponding to three risk dimensions: traffic accident risk dimension, vehicle failure risk dimension and traffic rules violation risk dimension.

[0022] For the risk control decision unit of the traffic accident risk dimension, it contains more network sub-units due to the high difficulty of predicting traffic accident risks. After receiving the risk control description knowledge features generated previously, the risk control decision unit begins to predict risks. For example, based on the information that Mr. Zhang's vehicle stays in congested areas for a long time and drives during peak hours, combined with a large amount of historical data and algorithm models, the unit predicts that Mr. Zhang's risk probability of a traffic accident in the next week is 10%. This risk prediction data is obtained through a comprehensive analysis of multiple factors such as Mr. Zhang's frequent changes in driving speed, many steering operations (which may increase the risk of collision), and low driving fatigue (reducing risk).

[0023] For the risk control decision unit of vehicle failure risk dimension, its prediction difficulty is relatively low, and the number of network sub-units is relatively small. Based on Mr. Zhang’s driving habits (a high proportion of light braking and accelerator pedaling, relatively small wear and tear on vehicle components) and the road conditions where the vehicle often drives (although there is congestion, the road conditions are good), the unit predicts that the risk probability of Mr. Zhang’s vehicle failure in the next month is 5%.

[0024] For the risk control decision-making unit of the risk dimension of violating traffic rules, considering the moderate difficulty of prediction, the number of its network sub-units is also at an intermediate level. Combining factors such as Mr. Zhang’s driving time pattern (driving during peak hours) and driving operation habits (low frequency of operating errors), it is predicted that the risk probability of him violating traffic rules in the next month is 3%.

[0025] This target risk control decision network is generated by constructing a sample learning sequence through a large amount of prior IoV user data, and obtaining the risk control profile feature data of each risk triggering event associated with each sample IoV user data in the sample learning sequence for joint network parameter learning. For example, data from 1,000 different users are collected, and each user's data contains information such as their driving status, driving behavior, and whether a traffic accident, vehicle failure, or violation of traffic rules has occurred. Based on this data, a sample learning sequence is constructed, and then the training cost layers corresponding to different risk control decision units are used to perform cyclic network parameter learning, and finally this target risk control decision network that can accurately predict risks is generated.

[0026] Step S130, obtaining the risk control management requirements of the target Internet of Vehicles user data, and the priori risk control portrait feature data of multiple reference risk control response strategies in the multiple risk dimensions respectively.

[0027] Continuing with Mr. Zhang as an example, the Internet of Vehicles service provider determined his risk control management needs based on factors such as Mr. Zhang's vehicle usage and insurance purchase. For example, since Mr. Zhang's vehicle is relatively new and has purchased comprehensive insurance, the risk control management needs of the Internet of Vehicles service provider are to minimize unnecessary risk control costs while ensuring service quality, while ensuring Mr. Zhang's driving safety and compliance with traffic rules.

[0028] For multiple reference risk control response strategies, assume that there are three:

[0029] Strategy A: This is a relatively conservative risk control strategy. In terms of traffic accident risk, its prior risk control profile feature data shows that for high-risk users (traffic accident risk probability exceeding 15%), it will actively push safe driving reminder information, such as real-time road conditions, safe driving tips, etc., and conduct a detailed analysis of the user's driving behavior once a week and provide improvement suggestions; in terms of vehicle failure risk, for users with a risk probability exceeding 8%, it will regularly remind them to maintain their vehicles and recommend nearby high-quality repair shops; in terms of traffic rule violation risk, for users with a risk probability exceeding 5%, it will push traffic rule interpretations and violation case warnings.

[0030] Strategy B: This is a moderate risk control strategy. In terms of traffic accident risk, for users with a risk probability between 8% and 15%, a safe driving reminder is pushed every two weeks; in terms of vehicle failure risk, for users with a risk probability between 5% and 8%, a vehicle maintenance reminder is pushed every two months; in terms of traffic rule violation risk, for users with a risk probability between 3% and 5%, a traffic rule interpretation is pushed once a month.

[0031] Strategy C: This is a relatively loose risk control strategy. In terms of traffic accident risk, only users with a risk probability of more than 20% are given irregular safe driving reminders; in terms of vehicle failure risk, only users with a risk probability of more than 10% are reminded to perform vehicle maintenance; in terms of traffic rule violation risk, relevant information is rarely pushed proactively unless the user's risk probability exceeds 8%.

[0032] The prior risk control profile feature data of these reference risk control response strategies in different risk dimensions are obtained through the analysis and summary of a large amount of historical user data. For example, statistical analysis is conducted on the effects of different strategies implemented by the past 10,000 users, which users have reduced the risk of traffic accidents after accepting a certain strategy, and which users have effectively controlled the risk of vehicle failure after accepting another strategy, etc. Based on these data, the prior risk control profile feature data of each strategy in different risk dimensions are constructed.

[0033] Step S140, based on the risk prediction data of the target risk dimension corresponding to the risk control management demand and the obtained priori risk control portrait feature data, generate the risk control response matching degree of the multiple reference risk control response strategies for the target Internet of Vehicles user data in the target risk dimension.

[0034] For example, based on Mr. Zhang's situation, assuming that the current target risk dimension is the traffic accident risk dimension, Mr. Zhang's traffic accident risk prediction data is 10%.

[0035] For strategy A, we analyze it from multiple dimensions:

[0036] In the risk factor dimension, the risk factors in the risk prediction data, such as frequent speed changes and driving during peak hours, are mapped and analyzed with the response plans for these risk factors in strategy A (such as pushing real-time traffic information and safe driving tips). It is found that these response plans can better deal with Mr. Zhang's risk factors, because real-time traffic information can help him plan his route in advance and reduce sudden changes in speed, and safe driving tips can improve his driving safety during peak hours. According to the importance weights of risk factors (such as the weight of frequent speed changes is 0.4, the weight of driving during peak hours is 0.3, etc.) and the ability of response plans to suppress or alleviate risk factors (such as the ability of pushing real-time traffic information to alleviate frequent speed changes is 0.6, and the ability of safe driving tips to alleviate driving during peak hours is 0.5, etc.), a quantitative risk factor response effectiveness evaluation index is constructed.

[0037] In the risk trend dimension, the risk trend in the risk prediction data (for example, as Mr. Zhang's driving time increases, the risk of traffic accidents may gradually increase) is analyzed in response to the risk trend corresponding to the historical response in Strategy A. It is found that the weekly driving behavior analysis and improvement suggestions of Strategy A can better adapt to this risk trend because problems in driving habits can be discovered and improved in a timely manner. A rule-based evaluation algorithm is constructed using the change amplitude and frequency of risk trends, as well as the adjustment speed and effect of Strategy A under different risk trends, to generate quantitative risk trend adaptability evaluation indicators.

[0038] In the resource utilization dimension, we compared the resource input information represented by the risk prediction data (for example, the network resources and server resources required for the transmission and processing of Mr. Zhang's vehicle data) with the resource allocation mode in Strategy A (such as the server bandwidth and storage resources required for pushing information), and evaluated the performance of different reference risk control response strategies in the resource utilization efficiency dimension. We found that Strategy A had relatively low resource utilization efficiency because it pushed more information every week.

[0039] In terms of execution process, we analyze the matching degree between the risk scenarios involved in the risk prediction data (such as traffic accident scenarios on urban roads) and the risk control measures execution process in Strategy A (such as pushing road condition information first and then analyzing driving behavior). We found that the overall matching degree is high, because road condition information can provide timely assistance during driving, and driving behavior analysis can follow up to improve driving habits.

[0040] According to the above multi-dimensional correlation analysis result matrix, each evaluation index in the quantitative evaluation index system is generated. Then, the hierarchical structure model is constructed using the hierarchical analysis method, taking the Internet of Vehicles environmental factors (such as network conditions, data transmission stability), risk dimension contribution values ​​(such as the importance of traffic accident risk in the overall risk) as the criterion layer, and each evaluation index as the solution layer. The relative importance between the elements of each layer is determined, and the comprehensive weight vector of each evaluation index is calculated. Finally, the value of strategy A on each quantitative evaluation index is multiplied by the weight value in the corresponding comprehensive weight vector to obtain the weighted index value, and all weighted index values ​​are summed to obtain the comprehensive evaluation value. This comprehensive evaluation value is the risk control response matching degree of strategy A for Mr. Zhang's Internet of Vehicles user data in the traffic accident risk dimension.

[0041] The same method is used to analyze strategies B and C to obtain their matching degree of risk control response for Mr. Zhang in the dimension of traffic accident risk.

[0042] Step S150: determining at least one target risk control response strategy for the target Internet of Vehicles user data according to the risk control response matching degree.

[0043] Assume that after calculation in step S140, it is obtained that the matching degree of strategy A for Mr. Zhang's risk control response in the traffic accident risk dimension is 0.7, the matching degree of strategy B is 0.6, and the matching degree of strategy C is 0.5.

[0044] According to the matching degree of these risk control responses, since Strategy A has the highest matching degree, Strategy A is determined as Mr. Zhang’s target risk control response strategy in the traffic accident risk dimension. This means that the Internet of Vehicles service provider will manage Mr. Zhang’s traffic accident risk in accordance with Strategy A, such as actively pushing safe driving reminder information, conducting detailed analysis of his driving behavior every week, and providing improvement suggestions.

[0045] If similar calculations and analyses are performed on the dimensions of vehicle failure risk and traffic rule violation risk, and different target risk control response strategies are determined, then the Internet of Vehicles service provider will integrate these target risk control response strategies to conduct comprehensive risk control management of Mr. Zhang's entire Internet of Vehicles user data to meet the risk control management needs determined previously, that is, while ensuring service quality, minimize unnecessary risk control costs, and at the same time ensure Mr. Zhang's driving safety and compliance with traffic rules.

[0046] Based on the above steps, the embodiment of the present application generates accurate risk control description knowledge features by comprehensively considering the dynamic driving state characteristics and risk association characteristics of the target Internet of Vehicles user, thereby achieving a comprehensive and detailed characterization of the user's driving risk. Utilizing the constructed target risk control decision network, this method can generate accurate risk prediction data for multiple risk dimensions, wherein the design of the risk control decision unit and the dynamic adjustment mechanism of the number of network sub-units ensure the accuracy and efficiency of risk prediction. Furthermore, by combining the user's risk control management needs and the prior risk control portrait feature data of the reference risk control response strategy, the present invention can accurately calculate the matching degree of each strategy with the target risk dimension, and tailor the most suitable risk control response strategy for the user. This method not only improves the risk control accuracy and response speed in the Internet of Vehicles environment, but also significantly enhances the pertinence and effectiveness of the risk control strategy, providing a safer and more reliable driving environment for Internet of Vehicles users.

[0047] In a possible implementation, performing joint network parameter learning according to the training cost layers respectively corresponding to the multiple risk control decision units to generate the target risk control decision network includes:

[0048] Step A110, constructing a sample learning sequence based on multiple prior Internet of Vehicles user data with risk trigger times greater than a set number, and obtaining risk control portrait feature data of each risk triggering event associated with each sample Internet of Vehicles user data in the sample learning sequence.

[0049] Step A120, determining the sample risk control description knowledge points of the sample Internet of Vehicles user data in the multiple risk dimensions based on the risk control portrait feature data.

[0050] Step A130, through the sample learning sequence and each sample risk control description knowledge point, and the training cost layers corresponding to the multiple risk control decision units, the initialized risk control decision network is subjected to cyclic network parameter learning to generate the target risk control decision network.

[0051] In a possible implementation, the risk control decision network includes at least a first risk control knowledge mining unit, a second risk control knowledge mining unit, and a plurality of risk control decision units.

[0052] Step A131, in each network parameter learning stage of the risk control decision network, extract a sample dynamic driving state feature of a sample Internet of Vehicles user data according to the first risk control knowledge mining unit, and extract a sample risk association feature of the sample Internet of Vehicles user data according to the second risk control knowledge mining unit.

[0053] Step A132, obtaining sample risk control description knowledge characteristics based on the sample dynamic driving state characteristics and the sample risk association characteristics, and generating sample risk prediction data of the sample Internet of Vehicles user data in multiple risk dimensions based on the sample risk control description knowledge characteristics through the multiple risk control decision units.

[0054] Step A133, based on the training cost layers corresponding to the multiple risk control decision units, generates errors between the sample risk prediction data of the multiple risk dimensions and the corresponding sample risk control description knowledge points, and trains the risk control decision network based on the generated errors.

[0055] In this embodiment, taking the previously mentioned Internet of Vehicles system as an example, in order to construct the target risk control decision network, firstly, a sample learning sequence is constructed based on multiple prior Internet of Vehicles user data with risk trigger times greater than the set times. Assuming that the set number is 10 times, the prior Internet of Vehicles user data with traffic accident risk trigger times exceeding 10 times, vehicle failure risk trigger times exceeding 10 times, and traffic rule violation risk trigger times exceeding 10 times in the past year are screened out from the Internet of Vehicles database. For these user data, the risk control portrait feature data of each risk trigger event associated with each sample Internet of Vehicles user data is obtained. For example, for a user who often has traffic accidents, his risk control portrait feature data includes information such as vehicle speed, driving time, weather conditions, road type, and driver's braking operation habits at the time of the accident; for risk trigger events related to vehicle failures, the mileage before the vehicle failure, recent maintenance conditions, and wear data of parts and components will be recorded; for traffic rule violations, the type of violation (such as speeding, running a red light, etc.), the location and time of the violation, and other data will be included.

[0056] Based on the acquired risk control portrait feature data, the sample risk control description knowledge points of each sample Internet of Vehicles user data in multiple risk dimensions are determined. Taking the traffic accident risk dimension as an example, for a certain sample user, if his accidents mostly occur at night and on roads with high speeds, and the accident rate is extremely low during the day and on roads with moderate speeds, then based on this feature, the sample risk control description knowledge point of the traffic accident risk dimension is determined to be that the risk is higher when driving at high speeds at night. In the vehicle failure risk dimension, if the mileage of a vehicle exceeds 100,000 kilometers and has not been maintained recently, the failure risk increases significantly, which becomes the sample risk control description knowledge point of the sample in the vehicle failure risk dimension. For the traffic rule violation risk dimension, if a user violates the rules many times at intersections without traffic lights, then the sample risk control description knowledge point in this risk dimension is that it is easy to violate the rules at intersections without traffic lights.

[0057] The risk control decision network includes a first risk control knowledge mining unit, a second risk control knowledge mining unit and multiple risk control decision units. In each network parameter learning stage of the risk control decision network, the first risk control knowledge mining unit extracts the sample dynamic driving state characteristics for a sample Internet of Vehicles user data. For example, for Mr. Zhang's vehicle data, the first risk control knowledge mining unit will extract the average speed and speed fluctuation range of different sections from the speed sensor data; extract the maximum, minimum and change frequency of acceleration from the acceleration sensor data; obtain the frequency and angle change range of steering operations from the driving direction sensor data; combine the geographic location information to determine the proportion of driving time on different road types (such as highways, urban main roads, rural roads), etc., which constitute the sample dynamic driving state characteristics. At the same time, the second risk control knowledge mining unit extracts sample risk-related features. For example, it analyzes the distribution of brake and accelerator operation force from Mr. Zhang's operation data, and obtains the proportion and continuity characteristics of light and heavy pedaling. It calculates the proportion and standard deviation of driving time in different time periods (such as weekdays and weekends) from the driving time pattern, and counts the number of times the continuous driving time exceeds the set time threshold and the frequency of operating errors from the driving fatigue level characteristics.

[0058] The sample risk control description knowledge features are obtained based on the extracted sample dynamic driving state features and sample risk association features. For example, if Mr. Zhang drives under congested urban road conditions (sample dynamic driving state features) and brakes frequently and with unstable force (sample risk association features), the sample risk control description knowledge features formed by combining this information may indicate that there is a certain risk of unstable braking operations under congested road conditions. Then, through multiple risk control decision units, based on the sample risk control description knowledge features, sample risk prediction data of sample Internet of Vehicles user data in multiple risk dimensions are generated. For example, in the dimension of traffic accident risk, the risk control decision unit predicts the risk probability of Mr. Zhang having a traffic accident in the future based on the above-mentioned sample risk control description knowledge features and a large amount of historical data comparative analysis.

[0059] Based on the training cost layers corresponding to multiple risk control decision units, the errors between the sample risk prediction data of multiple risk dimensions and the corresponding sample risk control description knowledge points are generated. For example, in the dimension of traffic accident risk, if the sample risk control description knowledge point indicates that the accident risk is high under a certain driving state, but the risk probability given by the sample risk prediction data is low, a large error will occur. Based on these errors, the risk control decision network is trained. By continuously adjusting the network parameters to reduce the error, this process is repeated in multiple network parameter learning stages, and finally a target risk control decision network is generated, which can more accurately predict risks based on Internet of Vehicles user data.

[0060] In a possible implementation, step A131 includes:

[0061] Step A1311, analyze the structure and metadata information of the sample Internet of Vehicles user data, and identify the data part that directly or indirectly reflects the vehicle's driving road condition status according to the pre-set driving state feature identification rules, and generate a first data subset related to the dynamic driving state characteristics, wherein the data that directly reflects the driving road condition status includes the vehicle's speed, acceleration, and frequent changes in the driving direction, and the data that indirectly reflects the driving road condition status is data related to the geographic location, specifically including the traffic congestion index and road type distribution of the vehicle's area.

[0062] In this embodiment, taking the sample Internet of Vehicles user data of Mr. Zhang in the Internet of Vehicles system mentioned above as an example, the sample Internet of Vehicles user data contains data from many sources, such as data collected by the vehicle's own sensors, positioning data related to geographic location, and metadata information such as timestamps. According to the pre-set driving state feature identification rules, the data part that directly or indirectly reflects the vehicle's driving road condition state begins to be identified, thereby generating a first data subset related to the dynamic driving state feature. The data part that directly reflects the driving road condition state, such as the vehicle's speed information, can obtain the real-time speed value sequence of the vehicle during driving from the speed sensor data; acceleration information, that is, the change of acceleration value recorded by the acceleration sensor when the vehicle accelerates or decelerates, such as the change of acceleration when the vehicle starts, brakes, or drives under different road conditions; information on frequent changes in driving direction, which is obtained through the vehicle's steering sensor data, such as relevant data on the vehicle changing the driving direction multiple times in a short period of time. In terms of data that indirectly reflects the state of driving road conditions, taking the traffic congestion index and road type distribution of the area where the vehicle is located as an example, the vehicle's positioning system interacts with the traffic data platform to obtain the traffic congestion index of the area where the vehicle is located, and at the same time determines the distribution of road types on which the vehicle is traveling based on map data, such as the situation of driving on different types of roads such as urban main roads, branch roads or highways. These data together constitute the first data subset related to dynamic driving state characteristics.

[0063] Step A1312: Use the first risk control knowledge mining unit to clean the input first data subset, standardize the cleaned first data subset, and then sort the standardized first data subset in chronological order or logical order to generate a first target data subset.

[0064] During the data cleaning process, outliers and erroneous data are removed from the data. For example, individual extremely large or extremely small speed values ​​that may be caused by sensor failure, these data that do not conform to normal driving logic will be identified and removed. Then the first data subset after cleaning is standardized to make data of different ranges and magnitudes comparable. For example, the speed data is unified into a specific unit and value range, and the acceleration data is also standardized accordingly. Then, the standardized first data subset is sorted in chronological order or logical order, such as sorting the speed, acceleration, driving direction and other data in chronological order during the vehicle's driving process to generate the first target data subset.

[0065] Step A1313, perform feature extraction operations on the first target data subset through the first risk control knowledge mining unit to construct sample dynamic driving state features of sample Internet of Vehicles user data, wherein for speed-related data, extract statistical features of the mean, variance, maximum and minimum values ​​of the speed, and for acceleration data, calculate the characteristics of the frequency of acceleration change and the frequency of occurrence of the peak value of acceleration, and obtain the characteristics of the angle and frequency of change of the driving direction from the driving direction data, and extract the proportion of the vehicle's driving time on different road types and the residence time in congested areas in combination with the geographic location-related data, and summarize them as sample dynamic driving state features.

[0066] For speed-related data, extract the statistical features of the mean, variance, maximum, and minimum values ​​of the speed. Take Mr. Zhang's vehicle as an example. Assume that after processing the speed data during a period of driving, the mean is 40 kilometers per hour, which indicates his average driving speed during this period of time. The variance reflects the degree of speed fluctuation. If the variance is small, it means that the driving speed is relatively stable. The maximum value may be 60 kilometers per hour, which is his fastest speed during this period of time. The minimum value is 20 kilometers per hour, which may be the speed when encountering traffic congestion or speed bumps. For acceleration data, calculate the characteristics of the frequency of acceleration change and the frequency of peak acceleration. For example, in traffic conditions with frequent start-stop, the frequency of acceleration change is high, while the frequency of change is low during stable driving. The frequency of peak acceleration reflects the frequency of sudden acceleration or braking of the vehicle. From the driving direction data, obtain the characteristics of the angle and frequency of change of the driving direction. If Mr. Zhang turns frequently during driving, the frequency of driving direction changes will be high, and the angle of change at each turn can also reflect the characteristics of his driving trajectory. Combined with the geographic location-related data, extract the characteristics of the vehicle's driving time ratio on different road types and the time it stays in congested areas. Assume that Mr. Zhang's total driving time on urban roads is 1 hour, of which 30 minutes are spent on main roads and 20 minutes on branch roads. Then the driving time on main roads and branch roads is 50% and 33.3% respectively. If he stays in a congested area for 15 minutes, this data reflects his situation under congested road conditions. These features are summarized to constitute the dynamic driving state features of the sample.

[0067] Step A1314, based on the predefined risk-related feature identification rules, identify the data portion related to the multi-dimensional driving behavior characteristics in the sample Internet of Vehicles user data, and construct a second data subset related to the risk-related features, wherein the multi-dimensional driving behavior characteristics include the driver's operating habit characteristics, driving time regularity characteristics, and driving fatigue level characteristics.

[0068] Multi-dimensional driving behavior characteristics include the driver's operating habit characteristics, driving time regularity characteristics, and driving fatigue characteristics. For example, data related to the driver's operating habit characteristics are identified from the vehicle's operating record data, such as the brake and accelerator pedal pedaling data; data related to driving time regularity characteristics are obtained from the driving time records, such as the start and end time of each driving; data related to driving fatigue characteristics are inferred from the driver's physiological state monitoring data (if any) or based on indirect data such as driving time. These data together constitute the second data subset.

[0069] Step A1315: After the second risk control knowledge mining unit performs data quality assessment and correction on the input second data subset, a data transformation operation is performed on the corrected second data subset to generate a second target data subset.

[0070] During the data quality assessment process, the integrity and accuracy of the data are checked. For example, whether there are missing values ​​in the brake and accelerator operation data, if so, they need to be supplemented or marked; for inaccurate data, such as the deviation of the operation force data caused by sensor error, correction is performed. Then, the corrected second data subset is subjected to data transformation operations, such as converting the time data into a specific format or normalizing the operation force data, etc., to generate the second target data subset.

[0071] Step A1316, extracting features from the second target data subset through the second risk control knowledge mining unit, and constructing the sample risk association features of the sample Internet of Vehicles user data, wherein the force distribution features of the brake and accelerator operations are specifically extracted from the driver's operating habit features, the force distribution features include the ratio of light and heavy steps, and the continuity features of the brake and accelerator operations, the continuity features include the duration of continuous light or heavy steps, and, for the driving time regularity features, calculate the proportion of driving time in different time periods and the standard deviation of driving time, and extract the number of times the continuous driving time exceeds the set time threshold and the frequency of operating errors during driving from the driving fatigue level features, and summarize them as the sample risk association features.

[0072] For example, the proportion of light and heavy braking and accelerator pedals is counted. Assume that Mr. Zhang's light braking and accelerator pedals account for 70% and heavy braking respectively, and the proportion of heavy braking and accelerator pedals account for 30%; and the continuity characteristics of braking and accelerator operation, that is, the duration of continuous light or heavy braking. If the longest duration of his continuous light braking is 5 seconds, this data reflects the continuity of his operation. For the driving time regularity characteristics, calculate the proportion of driving time in different time periods and the standard deviation of driving time. For example, Mr. Zhang's driving time from 7 to 9 in the morning accounts for 20% of the total driving time. By calculating the standard deviation, we can understand the fluctuation of his driving time. If the standard deviation is small, it means that his driving time is relatively regular. Extract the number of times the continuous driving time exceeds the set time threshold and the frequency of operation errors during driving from the driving fatigue level characteristics. Assuming that the set time threshold is 2 hours, Mr. Zhang's continuous driving time exceeds 2 hours 3 times, and the frequency of operation errors in the past 100 driving operations is 1%. These data are summarized to construct the sample risk association characteristics. Through such a series of rigorous operations, the sample dynamic driving state characteristics and sample risk association characteristics were accurately extracted from the sample Internet of Vehicles user data, providing an important basis for subsequent risk assessment and decision-making.

[0073] In a possible implementation manner, the step of determining the number of network sub-units of each risk control decision unit includes:

[0074] Step B110, for the risk prediction request in the risk prediction requests, determine the request distinguishing label set for the request attribute label of the first risk prediction request. The first risk prediction request is any one of the risk prediction requests, and the request distinguishing label corresponding to the first risk prediction request reflects: the distinguishing parameters between multiple risk prediction requests in the risk prediction request sequence corresponding to the first risk prediction request.

[0075] Step B120: configuring the number of network sub-units of the risk control decision unit corresponding to the first risk prediction request according to the prediction difficulty corresponding to the request distinguishing label.

[0076] In this embodiment, the steps for determining the number of network sub-units of each risk control decision unit are explained by taking the risk prediction requests of traffic accident risk prediction, vehicle failure risk prediction and traffic rule violation risk prediction mentioned above as examples.

[0077] First, for each risk prediction request, such as the first risk prediction request for traffic accident risk prediction, it is necessary to determine the request distinguishing label set by its request attribute label. In the risk prediction request sequence corresponding to the traffic accident risk prediction request, there are multiple risk prediction requests for different situations related to traffic accident risks, and these requests have distinguishing parameters. For example, from the perspective of environmental factors for accident occurrence, distinguishing parameters may include traffic accident risk prediction during the day and night, and under different weather conditions (sunny, rainy, foggy, etc.); from the perspective of road type, there are also differences in traffic accident risk prediction on highways, urban roads, and rural roads; from the perspective of driver behavior, traffic accident risk prediction under normal driving and fatigue driving conditions is also different. These different factors are combined to form the request distinguishing label of the traffic accident risk prediction request. This request distinguishing label reflects the distinguishing parameters between each sub-request in the traffic accident risk prediction request sequence, that is, the uniqueness of traffic accident risk prediction under different situations or conditions.

[0078] Then, according to the prediction difficulty corresponding to the request distinguishing label, the number of network sub-units of the risk control decision unit corresponding to the first risk prediction request is configured. Taking traffic accident risk prediction as an example, if the request distinguishing label of the traffic accident risk prediction request contains a combination of multiple complex environmental factors, driver behavior factors and road type factors, then its prediction difficulty is relatively high. For example, under complex weather conditions (such as heavy rain and dense fog), the prediction of traffic accident risks under different road types (such as roads with many bends in mountainous areas) combined with driver fatigue driving state requires consideration of many variables and complex interactions. In this case, in order to accurately predict the risk, it is necessary to configure more network sub-units to the corresponding risk control decision unit for traffic accident risk prediction. These network sub-units can handle different combinations of factors respectively. For example, some sub-units specialize in the impact of weather factors on accident risks, and other sub-units handle the risk situation under the interaction of road type and driver behavior.

[0079] Similarly, for vehicle failure risk prediction requests, the determination of the request discriminative label should take into account the discriminative parameters of the failure risk prediction of different components of the vehicle. For example, the differences between the failure risk predictions of different components such as engine failure, brake system failure, tire failure, and the impact of factors such as vehicle mileage, service life, and maintenance records on the failure risk prediction of different components. If the request discriminative label of the vehicle failure risk prediction request involves a complex combination of multiple component failure risks and the interaction of many influencing factors, the difficulty of prediction will increase. For example, considering the engine and brake system failure risk prediction of vehicles with high mileage and long-term unmaintained vehicles at the same time, more network sub-units are required to build the corresponding risk control decision-making units. These network sub-units can analyze specific influencing factors of different component failures, such as some sub-units focus on analyzing the impact of mileage on engine failure risk, and other sub-units analyze the impact of maintenance records on brake system failure risk, etc.

[0080] For the risk prediction request for violation of traffic rules, the determination of the request distinguishing label should take into account the distinguishing parameters of different types of traffic rules (such as speed limits, traffic signal rules, etc.), differences in traffic rules in different regions, and driver behavior habits on violation risk prediction. If the request distinguishing label of the traffic rule violation risk prediction request contains the combined influence of multiple complex traffic rule types, regional differences, and driver behavior habits, the prediction difficulty will increase. For example, in different regions (some regions have special traffic rules) and drivers' special driving habits (such as drivers in some regions have poor compliance with speed limits), more network sub-units are needed to build the corresponding risk control decision-making units. These network sub-units can handle the differences in traffic rules in different regions, the risk of violation of different types of traffic rules, and the relationship between driver behavior habits and violation risks. By configuring the number of network sub-units of the risk control decision-making unit according to the prediction difficulty corresponding to the request distinguishing label of each risk prediction request, a more accurate and efficient risk control decision network can be built, so as to more effectively predict the risk of vehicle network user data.

[0081] In a possible implementation, step A120 includes:

[0082] For each risk dimension, for each sample Internet of Vehicles user data, the risk triggering probability of the current sample Internet of Vehicles user data in the current risk dimension is generated based on the risk control portrait feature data of each risk triggering event associated with the current sample Internet of Vehicles user data.

[0083] Obtain the risk triggering mean probability of all prior IoV user data in the current risk dimension under the vehicle driving condition corresponding to the current sample IoV user data.

[0084] If the risk trigger probability corresponding to the current risk dimension of the current sample Internet of Vehicles user data is greater than the risk trigger mean probability, then the sample risk control description knowledge point of the current sample Internet of Vehicles user data is determined to be a positive risk control description knowledge point; otherwise, the sample risk control description knowledge point of the current sample Internet of Vehicles user data is determined to be a negative risk control description knowledge point.

[0085] In a possible implementation, step A120 may further include:

[0086] For each risk dimension, the risk triggering probability of the current sample Internet of Vehicles user data in the current risk dimension is generated based on the risk control portrait feature data of each risk triggering event associated with the current sample Internet of Vehicles user data.

[0087] Obtain the past risk triggering probabilities of all similar sample Internet of Vehicles user data under the vehicle driving condition corresponding to the current sample Internet of Vehicles user data, and divide the similar sample Internet of Vehicles user data into multiple risk levels according to the set allocation strategy.

[0088] The target risk level corresponding to the risk trigger probability corresponding to the current sample Internet of Vehicles user data in the current risk dimension is used as a sample risk control description knowledge point of the current sample Internet of Vehicles user data in the current risk dimension.

[0089] In the risk management system of the Internet of Vehicles, taking the traffic accident risk dimension, vehicle failure risk dimension and traffic rule violation risk dimension as examples, the paper elaborates in detail the process of determining the sample risk control description knowledge points of each sample Internet of Vehicles user data in multiple risk dimensions based on the feature data of each risk control portrait.

[0090] First, for the traffic accident risk dimension, for each sample IoV user data, the risk triggering probability of the current sample IoV user data in the traffic accident risk dimension is generated based on the risk control portrait feature data of each risk triggering event associated with the current sample IoV user data. For example, for Mr. Zhang's sample IoV user data, its risk control portrait feature data includes information such as his driving speed changes, driving time distribution, driving road type, weather conditions, and brake and accelerator operation habits. Assume that through analysis, it is found that Mr. Zhang drives at night at a higher rate, and the speed fluctuates greatly when driving at night. The frequency of driving on complex roads (such as mountain roads with many bends) is also high. Combining historical data and a specific risk assessment algorithm, it is concluded that Mr. Zhang's risk triggering probability in the traffic accident risk dimension is 10%.

[0091] Next, the mean probability of risk triggering of all prior Internet of Vehicles user data in the traffic accident risk dimension under the vehicle driving road condition corresponding to the current sample Internet of Vehicles user data is obtained. For example, under vehicle driving road conditions such as the types of roads that Mr. Zhang often drives on (including urban roads and some mountain roads) and similar driving time distributions (such as morning and evening rush hours on weekdays), the mean probability of traffic accident risk triggering is 8% as calculated from a large amount of prior Internet of Vehicles user data. Since the risk trigger probability of 10% corresponding to Mr. Zhang in the traffic accident risk dimension is greater than the risk trigger mean probability of 8%, according to the first implementation method, the sample risk control description knowledge point of Mr. Zhang's sample Internet of Vehicles user data in the traffic accident risk dimension is determined to be a positive risk control description knowledge point. This means that in terms of traffic accident risk, Mr. Zhang needs more active risk prevention and control measures than other users at the average level, and may need more frequent safety reminders or driving behavior analysis.

[0092] Then, the second implementation method is used for explanation. Similarly, for the traffic accident risk dimension, for Mr. Zhang’s sample Internet of Vehicles user data, the risk trigger probability in the traffic accident risk dimension is generated based on the risk control portrait feature data of each associated risk triggering event, which is 10%. Then, the past risk trigger probability of all similar sample Internet of Vehicles user data under the driving road condition of Mr. Zhang’s vehicle is obtained. Assume that in the similar sample Internet of Vehicles user data that drives the same type of vehicle as Mr. Zhang, drives on similar roads (such as urban main roads and mountain roads), and has a similar driving time distribution (such as commuting time on weekdays), the past risk trigger probability distribution is as follows: 30% of users have a risk trigger probability below 5%, 40% of users are between 5% and 10%, 20% of users are between 10% and 15%, and 10% of users are above 15%. According to the set allocation strategy, these similar sample Internet of Vehicles user data are divided into multiple risk levels, for example, the risk trigger probability below 5% is divided into a low risk level, 5% - 10% is divided into a medium risk level, 10% - 15% is divided into a higher risk level, and more than 15% is divided into a high risk level. Since Mr. Zhang's risk trigger probability is 10%, the corresponding target risk level is a higher risk level, so this higher risk level is used as a sample risk control description knowledge point for Mr. Zhang's sample Internet of Vehicles user data in the traffic accident risk dimension. This shows that in comparison with similar users, Mr. Zhang is at a higher risk level, and measures for this risk level need to be considered when formulating risk prevention and control strategies, such as providing more targeted driving safety training or more detailed monitoring of his driving behavior.

[0093] For the vehicle failure risk dimension, the first implementation method is used. Analyze a sample Internet of Vehicles user data, whose risk control portrait feature data includes information such as the age of the vehicle, mileage, recent maintenance records, parts wear data, and driving habits (such as the frequency of sudden acceleration and sudden braking). Assume that through the analysis of these data and the combination with the vehicle failure risk assessment model, it is concluded that the risk trigger probability of the sample user in the vehicle failure risk dimension is 12%. Under the vehicle driving road conditions corresponding to the sample user (such as frequent driving on roads with poor road conditions, large changes in temperature and humidity in the driving environment, etc.), the average probability of risk triggering of all prior Internet of Vehicles user data in the vehicle failure risk dimension is 10%. Because 12% is greater than 10%, it is determined that the sample risk control description knowledge point of the sample user in the vehicle failure risk dimension is a positive risk control description knowledge point, which means that the user needs more active risk prevention and control in terms of vehicle failure risk, such as more frequent vehicle inspection reminders or more detailed vehicle maintenance suggestions.

[0094] The vehicle failure risk dimension is described according to the second implementation method. For a certain sample user, the risk trigger probability in the vehicle failure risk dimension is 12% based on the risk control portrait feature data of the vehicle failure risk trigger event. The past risk trigger probability of all similar sample Internet of Vehicles user data under the vehicle driving road condition corresponding to the sample user is obtained. Assume that the past risk trigger probability distribution is: 20% of users with a risk trigger probability below 8%, 30% of users between 8% and 12%, 30% of users between 12% and 16%, and 20% of users above 16%. According to the set allocation strategy, these similar sample Internet of Vehicles user data are divided into low risk level (below 8%), medium risk level (8% - 12%), higher risk level (12% - 16%) and high risk level (above 16%). Since the risk trigger probability of the sample user is 12%, the corresponding target risk level is a higher risk level, so the higher risk level is used as the sample risk control description knowledge point of the sample user in the vehicle failure risk dimension. This means that among similar users, this user is at a higher risk level and needs to take corresponding levels of measures in terms of vehicle failure risk prevention and control, such as focusing on the inspection and maintenance of key vehicle components.

[0095] For the risk dimension of violating traffic rules, according to the first implementation method. Analyze the data of a sample Internet of Vehicles user, whose risk control portrait feature data includes the user's driving habits (such as whether he often speeds, whether he easily ignores traffic lights), driving time distribution (such as the frequency of driving in time periods with looser traffic control), and driving areas (such as driving in areas with stricter or looser traffic rules). Assume that through these data, it is concluded that the risk trigger probability of the sample user in the risk dimension of violating traffic rules is 8%. Under the vehicle driving road conditions corresponding to the sample user (such as often driving on urban roads and in areas with large traffic flow), the average probability of risk triggering of all prior Internet of Vehicles user data in the risk dimension of violating traffic rules is 6%. Because 8% is greater than 6%, it is determined that the sample risk control description knowledge point of the sample user in the risk dimension of violating traffic rules is a positive risk control description knowledge point, which indicates that the user needs more active prevention and control measures in terms of the risk of violating traffic rules, such as more frequent traffic rule reminders or education on driving behavior norms.

[0096] According to the second implementation method, the risk dimension of violating traffic rules is explained. For a certain sample user, based on the risk control portrait feature data of the risk triggering event of violating traffic rules, it is concluded that the risk trigger probability in the risk dimension of violating traffic rules is 8%. Obtain the past risk trigger probability of all similar sample Internet of Vehicles user data under the vehicle driving road condition corresponding to the sample user. Assume that the past risk trigger probability distribution is: users with a risk trigger probability below 5% account for 40%, users between 5% - 8% account for 30%, users between 8% - 10% account for 20%, and users above 10% account for 10%. According to the set allocation strategy, these similar sample Internet of Vehicles user data are divided into low risk level (below 5%), medium risk level (5% -8%), higher risk level (8% - 10%) and high risk level (above 10%). Since the risk trigger probability of the sample user is 8%, the corresponding target risk level is the medium risk level, so the medium risk level is used as the sample risk control description knowledge point of the sample user in the risk dimension of violating traffic rules. This means that among similar users, this user is at a medium-risk level and needs to take corresponding levels of measures to prevent and control the risk of violating traffic rules, such as appropriate traffic rules education and reminders.

[0097] By using these two methods to determine the sample risk control description knowledge points based on the characteristic data of each risk control portrait, it is possible to more accurately assess the risk of the sample Internet of Vehicles user data in various risk dimensions, thereby providing a more targeted basis for subsequent risk prevention and control decisions.

[0098] In a possible implementation, step A120 may further include:

[0099] For each risk dimension, target feature data directly related to the risk dimension is screened out from the risk control portrait feature data, and feature classification is performed on the target feature data under each risk dimension to generate static risk features and dynamic risk features.

[0100] Perform stability assessment on the static risk characteristics under each risk dimension to generate stability assessment results of the static risk characteristics under each risk dimension, and perform change trend analysis on the dynamic risk characteristics under each risk dimension to generate change trend analysis results of the dynamic risk characteristics under each risk dimension.

[0101] Based on the stability assessment results of static risk characteristics under each risk dimension and the change trend analysis results of dynamic risk characteristics, a comprehensive risk assessment is conducted on each risk dimension to generate an initial risk assessment value.

[0102] Based on the preliminary risk assessment values ​​of each risk dimension, a risk assessment framework containing assessment information of multiple risk dimensions is constructed.

[0103] Collect external factor information related to the Internet of Vehicles environment, and analyze the impact type judgment results of the external factor information on each risk dimension. According to the impact type judgment results of each external factor on each risk dimension, adjust the risk assessment framework to generate an adjusted risk assessment framework.

[0104] The assessment values ​​of each risk dimension in the adjusted risk assessment framework are normalized to generate a normalized risk assessment framework, and based on the assessment values ​​of each risk dimension in the normalized risk assessment framework and pre-set risk control description knowledge point determination rules, the sample risk control description knowledge points of each sample Internet of Vehicles user data in each risk dimension are determined.

[0105] In this embodiment, first, for each risk dimension, target feature data directly related to the risk dimension is screened out from the risk control portrait feature data, and feature classification is performed to generate static risk features and dynamic risk features.

[0106] For the traffic accident risk dimension, the risk control portrait feature data contains a lot of information, such as the driver's age, gender, driving experience, vehicle type, performance parameters, road type, traffic conditions (including traffic flow, traffic signal distribution, etc.) and weather conditions. Target feature data directly related to the traffic accident risk dimension may include the driver's driving experience, the vehicle's braking performance, the number and curvature of bends on the road, traffic flow, and weather conditions. These target feature data are classified by features. For example, the driver's age and vehicle's braking performance are regarded as static risk features, because these features are relatively stable in the short term and do not change frequently. Traffic flow and weather conditions are regarded as dynamic risk features, because traffic flow changes significantly in different time periods, and weather conditions may change at any time.

[0107] For the vehicle failure risk dimension, the risk control profile feature data may cover information such as the vehicle's age, mileage, maintenance records, parts wear, driving habits (such as frequency of sudden acceleration and braking), etc. The target feature data directly related to the vehicle failure risk dimension include the vehicle's age, mileage, parts wear, etc. The age and model of the vehicle are classified as static risk features because they remain basically unchanged for a long time; while the increase in mileage and changes in the degree of wear of parts are classified as dynamic risk features, which will continue to change with the use of the vehicle.

[0108] For the risk dimension of violating traffic rules, the risk control profile feature data includes information such as the driver's driving habits (such as whether he is prone to speeding, whether he obeys traffic signals), driving time distribution (such as driving during time periods with loose or strict traffic control), driving area (such as driving in areas with strict or loose traffic rules), etc. The selected target feature data directly related to this risk dimension include the driver's driving habits, the enforcement of traffic rules in the driving area, etc. The driver's basic driving habits (such as the long-term tendency to speed) are classified as static risk features, while temporary adjustments to traffic rules in the driving area (such as temporary changes to traffic rules on certain sections of the road due to special activities) and changes in traffic control intensity in different time periods are regarded as dynamic risk features.

[0109] Next, a stability assessment is performed on the static risk characteristics under each risk dimension to generate the stability assessment results of the static risk characteristics under each risk dimension, and a change trend analysis is performed on the dynamic risk characteristics under each risk dimension to generate the change trend analysis results of the dynamic risk characteristics under each risk dimension.

[0110] In the dimension of traffic accident risk, for static risk characteristics, such as the age of the driver, it is assumed that through statistical analysis, it is found that the probability of a traffic accident for drivers aged between 30 and 50 is relatively stable and low. This is the stability assessment result of the static risk characteristic of driver age. For the static risk characteristic of vehicle braking performance, if the vehicle is a newly purchased model equipped with an advanced braking system, its braking performance can remain stable and reliable for a long time. This is also a stability assessment result. For dynamic risk characteristics, taking traffic flow as an example, through the analysis of historical traffic flow data, it is found that the traffic flow during peak hours in the morning and evening on weekdays presents a regular peak, while the traffic flow distribution on holidays is relatively scattered and the overall flow is low. This is the change trend analysis result of the dynamic risk characteristic of traffic flow. In terms of weather conditions, according to the correlation analysis of meteorological data and traffic accident records, it is found that the risk of traffic accidents increases significantly in heavy rain and dense fog, and the frequency of such weather in specific seasons has a certain regularity. This is the change trend analysis result of the dynamic risk characteristic of weather conditions.

[0111] In the dimension of vehicle failure risk, for static risk characteristics, such as the age of the vehicle, through the statistics of failure data of a large number of vehicles of the same type, it is found that the failure risk of vehicles with a service life of 5-10 years is gradually increasing and relatively stable. This is the stability assessment result of the static risk characteristic of vehicle service life. For dynamic risk characteristics, in terms of the increase in mileage, according to the analysis of vehicle mileage data and failure records, it is found that for every increase of 10,000 kilometers, the wear risk of certain key components of the vehicle (such as tires, brake pads, etc.) will increase at a certain rate. This is the result of the change trend analysis of the dynamic risk characteristic of mileage.

[0112] In the risk dimension of violating traffic rules, for static risk characteristics, such as the driver's long-term tendency to speed, if the average number of speeding violations per month in the past year is relatively stable through analysis of historical driving records, this is the stability assessment result of the static risk characteristic of the driver's tendency to speed. For dynamic risk characteristics, the enforcement of traffic rules in the driving area, for example, during large-scale events in a certain area, traffic rules will be enforced more strictly, and the time schedule of the event is known in advance, this is the change trend analysis result of the dynamic risk characteristic of the enforcement of traffic rules in the driving area.

[0113] Based on the stability assessment results of static risk characteristics under each risk dimension and the change trend analysis results of dynamic risk characteristics, a comprehensive risk assessment is conducted on each risk dimension to generate an initial risk assessment value.

[0114] In the traffic accident risk dimension, combined with the stability assessment results of the driver's age (lower risk), the stability assessment results of the vehicle's braking performance (low risk), the trend analysis results of traffic flow changes (high risk during peak hours), and the trend analysis results of weather conditions (high risk in bad weather), the initial risk assessment value of the traffic accident risk dimension is generated through a pre-set risk assessment algorithm (such as assigning different weights to different factors and calculating them comprehensively). Assuming that in an evaluation system with a full score of 10, based on the comprehensive evaluation of the above factors, the initial risk assessment value of a sample Internet of Vehicles user in the traffic accident risk dimension is 4 points.

[0115] In the dimension of vehicle failure risk, based on the stability assessment results of the vehicle's service life (higher risk), the mileage change trend analysis results (the risk increases with mileage) and other factors, according to the risk assessment algorithm, the initial risk assessment value of a sample Internet of Vehicles user in the dimension of vehicle failure risk is 6 points.

[0116] In the risk dimension of violating traffic rules, taking into account factors such as the stability assessment results of the driver's speeding tendency (higher risk) and the changing trend analysis results of the enforcement of traffic rules in the driving area (high risk in special periods), the risk assessment algorithm is used to obtain an initial risk assessment value of 5 points for a sample Internet of Vehicles user in the risk dimension of violating traffic rules.

[0117] Based on the preliminary risk assessment values ​​of each risk dimension, a risk assessment framework containing assessment information of multiple risk dimensions is constructed. This risk assessment framework integrates the initial risk assessment values ​​of the traffic accident risk dimension, the vehicle failure risk dimension, and the traffic rule violation risk dimension, and may also include the weight information of different factors under each risk dimension and the risk assessment method. For example, this framework may be expressed in the form of a matrix, with rows representing different risk dimensions and columns representing the assessment values, factor weights, and other information under each risk dimension.

[0118] Collect external factor information related to the Internet of Vehicles environment, and analyze the impact type judgment results of the external factor information on each risk dimension. According to the impact type judgment results of each external factor on each risk dimension, adjust the risk assessment framework to generate an adjusted risk assessment framework.

[0119] External factors related to the Internet of Vehicles environment may include road infrastructure construction, new traffic regulations, and updates to Internet of Vehicles technology (such as upgrades to vehicle safety systems). In the traffic accident risk dimension, for example, new road infrastructure construction may improve road safety, which is a positive impact type judgment result. If new traffic regulations increase penalties for certain dangerous driving behaviors, this will also have a positive impact on traffic accident risk (promoting drivers to comply with traffic rules more). For the vehicle failure risk dimension, updates to Internet of Vehicles technology enable vehicles to monitor the health of components in real time, which is a positive impact type judgment result. For the traffic rule violation risk dimension, if the introduction of new traffic regulations increases the supervision of specific violations, this is a positive impact type judgment result. Based on these impact type judgment results, the risk assessment framework is adjusted. For example, in the traffic accident risk dimension, due to the positive impact of road infrastructure construction and new traffic regulations, the previously calculated initial risk assessment value may be reduced. The assessment value of the traffic accident risk dimension and other information are modified accordingly in the risk assessment framework to generate an adjusted risk assessment framework.

[0120] The assessment values ​​of each risk dimension in the adjusted risk assessment framework are normalized to generate a normalized risk assessment framework, and the sample risk control description knowledge points of each sample Internet of Vehicles user data in each risk dimension are determined based on the assessment values ​​of each risk dimension in the normalized risk assessment framework and the pre-set risk control description knowledge point determination rules.

[0121] Assume that in the adjusted risk assessment framework, the assessment value of the traffic accident risk dimension is 3 points, the assessment value of the vehicle failure risk dimension is 5 points, and the assessment value of the traffic rule violation risk dimension is 4 points. After normalization (for example, mapping the assessment value of each risk dimension to between 0 and 1), the normalized assessment value of the traffic accident risk dimension is 0.3, the normalized assessment value of the vehicle failure risk dimension is 0.5, and the normalized assessment value of the traffic rule violation risk dimension is 0.4. According to the pre-set risk control description knowledge point determination rules, for example, it is stipulated that the normalized assessment value between 0 and 0.3 is a low-risk sample risk control description knowledge point, between 0.3 and 0.6 is a medium-risk sample risk control description knowledge point, and between 0.6 and 1 is a high-risk sample risk control description knowledge point. Then for this sample IoV user data, the sample risk control description knowledge point in the traffic accident risk dimension is low risk, the sample risk control description knowledge point in the vehicle failure risk dimension is medium risk, and the sample risk control description knowledge point in the traffic rule violation risk dimension is medium risk. In this way, the process of determining the sample risk control description knowledge point of each sample IoV user data in each risk dimension based on the feature data of each risk control portrait is completed, providing a more accurate and comprehensive basis for subsequent risk management and control decisions.

[0122] In a possible implementation, after step A133, if the multiple risk control decision units all meet the training termination requirements, the network parameter learning is terminated and the target risk control decision network is generated. If there is a target risk control decision unit that does not meet the training termination requirements among the multiple risk control decision units, then according to the sample learning sequence and each sample risk control description knowledge point, and the training cost layer corresponding to the target risk control decision unit, when continuing to train the initialized risk control decision network for network parameter learning, the target risk control decision unit is trained, and the first risk control knowledge mining unit, the second risk control knowledge mining unit, and the neuron weight parameters of the risk control decision unit that meets the training termination requirements among the multiple risk control decision units are locked.

[0123] In this embodiment, the risk control decision-making units corresponding to traffic accident risks, vehicle failure risks, and traffic rule violation risks are taken as examples to illustrate this process.

[0124] Training termination requirements may be based on factors such as error convergence, number of training iterations, or achieving a predetermined risk prediction accuracy.

[0125] For the risk control decision unit corresponding to the traffic accident risk, it is assumed that the training termination requirement is that the error converges to a very small value, such as an error less than 0.01, and the risk prediction accuracy reaches more than 90%. During the training process, if the error of the risk control decision unit corresponding to the traffic accident risk continues to decrease and eventually reaches less than 0.01, and the risk prediction accuracy reaches 95%, this indicates that the risk control decision unit meets the training termination requirements. Similarly, similar checks are performed on the risk control decision units corresponding to vehicle failure risks and traffic rule violation risks.

[0126] If all risk control decision units, i.e., the risk control decision units corresponding to traffic accident risk, vehicle failure risk, and traffic rule violation risk, meet the training termination requirements, then the network parameter learning is terminated and the target risk control decision network is generated. This target risk control decision network can accurately predict the traffic accident risk, vehicle failure risk, and traffic rule violation risk of Internet of Vehicles user data.

[0127] However, if there is a target risk control decision unit among multiple risk control decision units that does not meet the training termination requirements, for example, the risk control decision unit corresponding to the traffic accident risk does not meet the training termination requirements, while the risk control decision units corresponding to the vehicle failure risk and the risk of violating traffic rules have met the requirements. At this time, according to the sample learning sequence and the knowledge points of the risk control description of each sample, as well as the training cost layer of the target risk control decision unit corresponding to the traffic accident risk, the initialized risk control decision network continues to train the network parameter learning. In this process, only the target risk control decision unit corresponding to the traffic accident risk is trained, and the neuron weight parameters of the first risk control knowledge mining unit, the second risk control knowledge mining unit, and the risk control decision units corresponding to the vehicle failure risk and the risk of violating traffic rules that have met the training termination requirements are locked.

[0128] The purpose of locking the neuron weight parameters of the risk control decision units that have met the requirements is to maintain the effectiveness of these units in the current state and avoid interference when training the target risk control decision units corresponding to the traffic accident risk. For example, the risk control decision unit corresponding to the vehicle failure risk has achieved an accurate prediction state through previous training, and its neuron weight parameters have been adjusted to appropriate values. If these parameters are not locked, the performance of the risk control decision unit corresponding to the vehicle failure risk may be reduced when the target risk control decision unit corresponding to the traffic accident risk is subsequently trained.

[0129] For the continued training of the target risk control decision unit corresponding to the traffic accident risk, the sample learning sequence provides a rich sample of Internet of Vehicles user data, and each sample risk control description knowledge point provides the correct reference standard for training. The training cost layer defines the cost function during the training process, which is used to measure the degree of difference between the predicted result and the correct result. By continuously adjusting the network parameters of the target risk control decision unit corresponding to the traffic accident risk, the error is gradually reduced until the training termination requirements are met. In this way, it can ensure that the entire risk control decision network can accurately predict risks in each risk dimension and provide accurate risk management services for Internet of Vehicles users.

[0130] In a possible implementation, step S140 includes:

[0131] Step S141, performing feature mapping and association analysis on the risk prediction data of the target risk dimension and the prior risk control portrait feature data of each reference risk control response strategy under the target risk dimension from multiple dimensions, and constructing a corresponding multi-dimensional association analysis result matrix, wherein the multi-dimensional association analysis result matrix is ​​used to reflect the relationship between the risk prediction data of the target risk dimension and the prior risk control portrait feature data of each reference risk control response strategy under the target risk dimension, wherein, in the risk factor dimension, various risk factors in the risk prediction data and the response plans for responding to the risk factors in the prior risk control portrait feature data are mapped and analyzed, in the risk trend dimension, the risk trend in the risk prediction data and the response results of the risk trend corresponding to the historical response in the prior risk control portrait feature data are analyzed, and, in the resource utilization dimension, the resource input information represented by the risk prediction data is compared with the resource allocation mode in the prior risk control portrait feature data, and the performance of different reference risk control response strategies in the resource utilization efficiency dimension is evaluated, and, in the execution process dimension, the matching degree between the risk occurrence scenario involved in the risk prediction data and the risk control measure execution process in the prior risk control portrait feature data is analyzed.

[0132] Step S142, based on the multi-dimensional correlation analysis result matrix, generate various evaluation indicators in the quantitative evaluation indicator system, specifically: based on the correlation analysis results of the risk factor dimension, for each risk factor and its corresponding response plan, combined with the importance weight of the risk factor and the ability of the response plan to suppress or alleviate the risk factor, construct a quantitative risk factor response effectiveness evaluation indicator, and, using the correlation analysis results of the risk trend dimension, quantify the risk trend adaptability performance of each reference risk control response strategy, and construct a rule-based evaluation algorithm based on the change amplitude and frequency of the risk trend and the adjustment speed and adjustment effect of the reference risk control response strategy under different risk trends, generate a quantitative risk trend adaptability evaluation indicator, and, based on the correlation analysis results of the resource utilization dimension, construct a quantitative resource utilization efficiency evaluation indicator, and, based on the correlation analysis results of the execution process dimension, construct an evaluation indicator for the rationality of the quantitative execution process.

[0133] Step S143, using the hierarchical analysis method to construct a hierarchical model, taking the Internet of Vehicles environmental factors, risk dimension contribution values, etc. as the criterion layer, and each evaluation indicator as the solution layer, determining the relative importance between the elements of each layer, and calculating the comprehensive weight vector of each evaluation indicator.

[0134] Step S144, according to the implementation of the quantitative evaluation index system and the comprehensive weight vector of each evaluation index, for each reference risk control response strategy, the value of the reference risk control response strategy on each quantitative evaluation index is multiplied by the weight value in the corresponding comprehensive weight vector to obtain a weighted index value, and all weighted index values ​​are summed to obtain a comprehensive evaluation value. The comprehensive evaluation value serves as the risk control response matching degree of each reference risk control response strategy for the target Internet of Vehicles user data in the target risk dimension, reflecting the matching degree of each reference risk control response strategy with the risk prediction data of the target risk dimension under multiple factors.

[0135] In this embodiment, first, feature mapping and correlation analysis are performed on the risk prediction data of the target risk dimension (traffic accident risk dimension) and the prior risk control portrait feature data of each reference risk control response strategy under the target risk dimension from multiple dimensions to construct a corresponding multi-dimensional correlation analysis result matrix.

[0136] In the risk factor dimension, the various risk factors in the risk prediction data include the driver's driving habits (such as the frequency of sudden acceleration and sudden braking), the type of road (such as the proportion of urban roads and highways), the distribution of driving time (such as the proportion of driving during peak hours), etc. For the reference risk control response strategy A, the response plan for the risk factors in its prior risk control portrait feature data may be to provide driving habit improvement suggestions for sudden acceleration and sudden braking behaviors, and to provide road condition warnings for driving during peak hours. The driving habit risk factors in the risk prediction data are mapped and analyzed with the driving habit improvement suggestions in the response plan. For example, if the risk prediction data shows that the driver has a high frequency of sudden acceleration and sudden braking, and the driving habit improvement suggestions in the response plan are highly targeted and there is data showing that the suggestions can effectively reduce such behaviors, then there is a high degree of correlation in this mapping relationship. Similarly, the risk factors of the road type are mapped and analyzed with the warning plans for different road types, and the driving time distribution risk factors are mapped and analyzed with the response plans for peak hours, etc.

[0137] In the risk trend dimension, the risk trend in the risk prediction data may be that as the driver ages, the risk of traffic accidents tends to gradually increase, or as the vehicle age increases, the risk of accidents will also change. The response results of the historical responses to the corresponding risk trends in the prior risk control portrait feature data of the reference risk control response strategy A are analyzed. If strategy A has a record of adjusting risk prevention and control measures in response to the increase in driver age or vehicle age, and these adjustments are shown in historical data to have a good inhibitory effect on risk trends, such as by adjusting the reminder frequency or providing different types of safety advice, then there will be better response results in the risk trend dimension.

[0138] In the resource utilization dimension, the resource input information represented by the risk prediction data may include the amount of Internet of Vehicles data transmission, data processing and computing resources required to obtain these risk prediction data. The resource allocation pattern in the prior risk control portrait feature data of the reference risk control response strategy A may be the server resources and network bandwidth required for data processing and push reminder information at a fixed time every week. Comparing the two, if the resource input required for the risk prediction data matches the resource allocation pattern of strategy A, that is, it will not cause excessive or insufficient use of resources, and the resource utilization efficiency is high while meeting the risk prevention and control requirements, then it will perform better in the resource utilization dimension.

[0139] In the execution process dimension, the risk scenarios involved in the risk prediction data include intersection accident risks on urban roads and rear-end collision risks on highways. The risk control measures execution process in the prior risk control portrait feature data of the reference risk control response strategy A may be to first conduct a key risk assessment on the intersection, then push relevant reminder information based on the assessment results, and then monitor driving behavior on the highway. Analyze the matching degree between the two. If the execution process is consistent with the risk occurrence scenario and can effectively respond to risks in various scenarios, then there is a high degree of matching in the execution process dimension. Through the analysis of the above four dimensions, a multi-dimensional correlation analysis result matrix is ​​constructed to comprehensively reflect the relationship between the risk prediction data of the target risk dimension and the prior risk control portrait feature data of the reference risk control response strategy A in the traffic accident risk dimension.

[0140] Next, based on the multi-dimensional correlation analysis result matrix, each evaluation index in the quantitative evaluation index system is generated.

[0141] Based on the correlation analysis results of the risk factor dimension, a quantitative risk factor response effectiveness evaluation index is constructed for each risk factor and its corresponding response plan. For example, for the sudden acceleration and braking behaviors in the driving habit risk factor, it is assumed that through a large amount of data statistical analysis, the importance weight of sudden acceleration and sudden braking behaviors to the risk of traffic accidents is 0.3 (indicating the degree of its contribution to the overall risk). Referring to the driving habit improvement suggestions in the risk control response strategy A, the inhibitory ability to reduce sudden acceleration and sudden braking behaviors is 0.6 (indicating that the plan can effectively reduce the probability of such behaviors), then the risk factor response effectiveness evaluation index is 0.3×0.6 = 0.18. Similarly, similar calculations are performed on the road type risk factor and the driving time distribution risk factor, and the overall response effectiveness evaluation index of strategy A in the risk factor dimension is obtained comprehensively.

[0142] The risk trend adaptability of each reference risk control response strategy is quantified using the correlation analysis results of the risk trend dimension. For example, the risk trend is that the increase in driver age leads to an increase in accident risk, with a change range of 5% per year and a significant change frequency every 5 years. The reference risk control response strategy A adjusts the prevention and control measures every 3 years under the risk trend of driver age growth, and the adjustment effect is that each adjustment can reduce the risk increase by 2%. A rule-based evaluation algorithm is constructed based on the change range and frequency of risk trends, as well as the adjustment speed and effect of strategy A under different risk trends. For example, a comprehensive calculation is performed based on the relationship between the adjustment speed and the frequency of risk changes, and the relationship between the adjustment effect and the risk change range, to generate a quantitative risk trend adaptability evaluation indicator.

[0143] Based on the results of the correlation analysis of the resource utilization dimension, a quantitative resource utilization efficiency evaluation index is constructed. For example, if the resource input required for risk prediction data is 100 units (including data transmission, computing resources, etc.), the actual resource usage under the resource allocation mode of the reference risk control response strategy A is 80 units, and under the risk prevention and control requirements, according to the relationship between resource input and output, the resource utilization efficiency is calculated to be 80%. This is the resource utilization efficiency evaluation index of strategy A.

[0144] Based on the correlation analysis results of the execution process dimension, an evaluation index for the rationality of the quantitative execution process is constructed. For example, for the intersection accident risk scenario in the risk prediction data, the proportion of the processing steps for intersection risks in the execution process of the reference risk control response strategy A accounts for 30% of the total execution process steps, and historical data verifies that this proportion can effectively deal with intersection risks. Based on the rationality standards of the execution process in different scenarios, the evaluation index of the rationality of the execution process is calculated.

[0145] Then, the hierarchical model is constructed using the analytic hierarchy process. The Internet of Vehicles environmental factors (such as network stability and data accuracy), risk dimension contribution value (such as the proportion of traffic accident risk in the overall risk), etc. are used as the criterion layer, and the various evaluation indicators such as risk factor response effectiveness evaluation index, risk trend adaptability evaluation index, resource utilization efficiency evaluation index, and execution process rationality evaluation index are used as the solution layer. Determine the relative importance of the elements in each layer. For example, through expert evaluation or historical data statistical analysis, determine that the importance ratio of network stability in the Internet of Vehicles environmental factors to the risk dimension contribution value is 0.6:0.4, and the importance ratio of risk factor response effectiveness evaluation index to risk trend adaptability evaluation index is 0.5:0.5, etc. Based on these relative importance relationships, calculate the comprehensive weight vector of each evaluation indicator.

[0146] Finally, according to the implementation of the quantitative evaluation index system and the comprehensive weight vector of each evaluation index, for each reference risk control response strategy (taking strategy A as an example), the value of strategy A on each quantitative evaluation index is multiplied by the weight value in the corresponding comprehensive weight vector to obtain the weighted index value. For example, the risk factor response effectiveness evaluation index value of strategy A is 0.18, and the corresponding weight value in the comprehensive weight vector is 0.2, then the weighted index value is 0.18×0.2 =0.036. All weighted index values ​​are summed to obtain the comprehensive evaluation value. This comprehensive evaluation value is the risk control response matching degree of the reference risk control response strategy A for the target Internet of Vehicles user data in the traffic accident risk dimension, which reflects the matching degree of strategy A with the risk prediction data in the traffic accident risk dimension under multiple factors. The same method can be used to calculate the risk control response matching degree of other reference risk control response strategies in the traffic accident risk dimension, thereby providing a basis for selecting the most suitable risk control response strategy.

[0147] Figure 2 The hardware structure of the risk control information processing system 100 based on the Internet of Vehicles provided by the embodiment of the present invention is shown as follows: Figure 2 As shown, the risk control information processing system 100 based on the Internet of Vehicles may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0148] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the Internet of Vehicles-based risk control information processing system 100 uses to execute or use to complete the exemplary method described in the present invention.

[0149] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the risk control information processing method based on the Internet of Vehicles in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0150] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned risk control information processing system 100 based on the Internet of Vehicles. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0151] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the above-mentioned risk control information processing method based on the Internet of Vehicles is implemented.

[0152] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, various features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A risk control information processing method based on Internet of Vehicles, characterized in that: The method comprises: Generate corresponding risk control description knowledge features based on the dynamic driving state features and risk association features of the target Internet of Vehicles user data; wherein the dynamic driving state features reflect the vehicle driving road condition state corresponding to the target Internet of Vehicles user data, and the risk association features reflect the multi-dimensional driving behavior features of the target Internet of Vehicles user data; Through the multiple risk control decision units included in the target risk control decision network, based on the risk control description knowledge characteristics, risk prediction data of the target Internet of Vehicles user data in multiple risk dimensions are generated; wherein, one risk control decision unit corresponds to a risk prediction request of one risk dimension, one risk control decision unit generates risk prediction data of one risk dimension, and the number of network subunits of each risk control decision unit is positively correlated with the prediction difficulty of the corresponding risk prediction request, and the target risk control decision network is generated by joint network parameter learning based on the training cost layers corresponding to the multiple risk control decision units respectively; Obtaining risk control management requirements for the target Internet of Vehicles user data, and prior risk control profile feature data of multiple reference risk control response strategies in the multiple risk dimensions; Generate the risk control response matching degree of the plurality of reference risk control response strategies for the target Internet of Vehicles user data in the target risk dimension according to the risk prediction data of the target risk dimension corresponding to the risk control management requirement and the obtained priori risk control portrait feature data; At least one target risk control response strategy for the target Internet of Vehicles user data is determined based on the risk control response matching degree.

2. The risk control information processing method based on the Internet of Vehicles according to claim 1 is characterized in that: Performing joint network parameter learning according to the training cost layers respectively corresponding to the multiple risk control decision units to generate the target risk control decision network includes: Constructing a sample learning sequence based on a plurality of prior Internet of Vehicles user data with a risk triggering frequency greater than a set frequency, and obtaining risk control portrait feature data of each risk triggering event associated with each sample Internet of Vehicles user data in the sample learning sequence; Determine, based on each risk control portrait feature data, sample risk control description knowledge points of each sample Internet of Vehicles user data in the multiple risk dimensions; Through the sample learning sequence and each sample risk control description knowledge point, as well as the training cost layers corresponding to the multiple risk control decision units, the initialized risk control decision network is subjected to cyclic network parameter learning to generate the target risk control decision network.

3. The risk control information processing method based on the Internet of Vehicles according to claim 2 is characterized in that: The risk control decision network at least includes a first risk control knowledge mining unit, a second risk control knowledge mining unit, and a plurality of risk control decision units; In each network parameter learning phase of the risk control decision network, extracting a sample dynamic driving state feature of a sample Internet of Vehicles user data according to the first risk control knowledge mining unit, and extracting a sample risk association feature of the sample Internet of Vehicles user data according to the second risk control knowledge mining unit; Obtaining sample risk control description knowledge features according to the sample dynamic driving state features and the sample risk association features, and generating sample risk prediction data of the sample Internet of Vehicles user data in multiple risk dimensions according to the sample risk control description knowledge features through the multiple risk control decision units; Generate errors between the sample risk prediction data of the multiple risk dimensions and the corresponding sample risk control description knowledge points according to the training cost layers respectively corresponding to the multiple risk control decision units, and train the risk control decision network according to the generated errors; The steps of extracting a sample dynamic driving state feature of a sample Internet of Vehicles user data according to the first risk control knowledge mining unit, and extracting a sample risk association feature of the sample Internet of Vehicles user data according to the second risk control knowledge mining unit include: Analyze the structure and metadata information of the sample Internet of Vehicles user data, and identify the data part that directly or indirectly reflects the vehicle's driving road condition status according to the preset driving state feature identification rules, and generate a first data subset related to the dynamic driving state feature, wherein the data that directly reflects the driving road condition status includes the vehicle's speed, acceleration, and frequent changes in the driving direction, and the data that indirectly reflects the driving road condition status is data related to the geographic location, specifically including the traffic congestion index and road type distribution of the area where the vehicle is located; Using the first risk control knowledge mining unit to perform data cleaning on the input first data subset, and after standardizing the cleaned first data subset, sorting the standardized first data subset in chronological order or logical order to generate a first target data subset; The first risk control knowledge mining unit performs a feature extraction operation on the first target data subset to construct a sample dynamic driving state feature of the sample Internet of Vehicles user data, wherein for speed-related data, statistical features of the mean, variance, maximum, and minimum values ​​of the speed are extracted; for acceleration data, features of the frequency of acceleration change and the frequency of occurrence of the peak value of acceleration are calculated; from the driving direction data, features of the angle of change of the driving direction and the frequency of change are obtained; and combined with the geographic location-related data, features of the proportion of the vehicle's driving time on different road types and the length of time it stays in a congested area are extracted, and the features are summarized as the sample dynamic driving state feature; According to a predefined risk-related feature identification rule, a data portion related to the multi-dimensional driving behavior feature in the sample Internet of Vehicles user data is identified, and a second data subset related to the risk-related feature is constructed, wherein the multi-dimensional driving behavior feature includes the driver's operating habit feature, driving time regularity feature, and driving fatigue degree feature; After the second risk control knowledge mining unit performs data quality assessment and correction on the input second data subset, a data transformation operation is performed on the corrected second data subset to generate a second target risk data subset; The second risk control knowledge mining unit is used to extract features from the second target data subset to construct sample risk association features of the sample Internet of Vehicles user data, wherein the force distribution features of brake and accelerator operations are specifically extracted from the driver's operating habit features, the force distribution features include the ratio of light and heavy steps, and the continuity features of brake and accelerator operations, the continuity features include the duration of continuous light or heavy steps, and, for the driving time regularity features, the proportion of driving time in different time periods and the standard deviation of driving time are calculated, and the number of times the continuous driving time exceeds the set time threshold and the frequency of operating errors during driving are extracted from the driving fatigue level features, which are summarized as the sample risk association features.

4. The method for processing risk control information based on the Internet of Vehicles according to any one of claims 1 to 3, characterized in that: The step of determining the number of network sub-units of each risk control decision unit includes: For the risk prediction request in each risk prediction request, determine a request distinguishing label set for the request attribute label corresponding to the first risk prediction request; wherein the first risk prediction request is any one of the risk prediction requests, and the request distinguishing label corresponding to the first risk prediction request reflects: in the risk prediction request sequence corresponding to the first risk prediction request, a distinguishing parameter between multiple risk prediction requests; According to the prediction difficulty corresponding to the request distinguishing label, the number of network sub-units of the risk control decision unit corresponding to the first risk prediction request is configured.

5. The risk control information processing method based on the Internet of Vehicles according to claim 2 is characterized in that: Determining the sample risk control description knowledge points of each sample Internet of Vehicles user data in the multiple risk dimensions based on each risk control portrait feature data includes: For each risk dimension, for each sample Internet of Vehicles user data, respectively, based on the risk control portrait feature data of each risk triggering event associated with the current sample Internet of Vehicles user data, generate the risk triggering probability of the current sample Internet of Vehicles user data in the current risk dimension; Obtain the risk triggering mean probability of all prior IoV user data in the current risk dimension under the vehicle driving condition corresponding to the current sample IoV user data; If the risk trigger probability corresponding to the current risk dimension of the current sample Internet of Vehicles user data is greater than the risk trigger mean probability, then the sample risk control description knowledge point of the current sample Internet of Vehicles user data is determined to be a positive risk control description knowledge point; otherwise, the sample risk control description knowledge point of the current sample Internet of Vehicles user data is determined to be a negative risk control description knowledge point.

6. The method for processing risk control information based on Internet of Vehicles according to claim 2, characterized in that: Determining the sample risk control description knowledge points of each sample Internet of Vehicles user data in the multiple risk dimensions based on each risk control portrait feature data includes: For each risk dimension, respectively, based on the risk control portrait feature data of each risk triggering event associated with the current sample Internet of Vehicles user data, generate the risk triggering probability of the current sample Internet of Vehicles user data in the current risk dimension; Obtaining the past risk triggering probabilities of all similar sample Internet of Vehicles user data under the vehicle driving road condition corresponding to the current sample Internet of Vehicles user data, and dividing the similar sample Internet of Vehicles user data into multiple risk levels according to a set allocation strategy; The target risk level corresponding to the risk trigger probability corresponding to the current sample Internet of Vehicles user data in the current risk dimension is used as a sample risk control description knowledge point of the current sample Internet of Vehicles user data in the current risk dimension.

7. The method for processing risk control information based on Internet of Vehicles according to claim 2, characterized in that: The step of determining, based on the risk control portrait feature data, sample risk control description knowledge points of each sample Internet of Vehicles user data in the multiple risk dimensions includes: For each risk dimension, target feature data directly related to the risk dimension is screened out from the risk control portrait feature data, and feature classification is performed on the target feature data under each risk dimension to generate static risk features and dynamic risk features; Conduct stability assessment on the static risk characteristics under each risk dimension, generate stability assessment results of the static risk characteristics under each risk dimension, and conduct change trend analysis on the dynamic risk characteristics under each risk dimension, generate change trend analysis results of the dynamic risk characteristics under each risk dimension; Based on the stability assessment results of static risk characteristics under each risk dimension and the change trend analysis results of dynamic risk characteristics, a comprehensive risk assessment is conducted on each risk dimension to generate an initial risk assessment value; Based on the preliminary risk assessment values ​​of each risk dimension, a risk assessment framework containing assessment information of multiple risk dimensions is constructed; Collecting external factor information related to the Internet of Vehicles environment, and analyzing the impact type judgment results of the external factor information on each risk dimension, adjusting the risk assessment framework according to the impact type judgment results of each external factor on each risk dimension, and generating an adjusted risk assessment framework; The assessment values ​​of each risk dimension in the adjusted risk assessment framework are normalized to generate a normalized risk assessment framework, and based on the assessment values ​​of each risk dimension in the normalized risk assessment framework and pre-set risk control description knowledge point determination rules, the sample risk control description knowledge points of each sample Internet of Vehicles user data in each risk dimension are determined.

8. The method for processing risk control information based on Internet of Vehicles according to claim 3, characterized in that: After training the risk control decision network based on the generated errors, the method further includes: If the multiple risk control decision units all meet the training termination requirements, the network parameter learning is terminated and the target risk control decision network is generated; If there is a target risk control decision unit among the multiple risk control decision units that does not meet the training termination requirements, then according to the sample learning sequence and each sample risk control description knowledge point, and the training cost layer corresponding to the target risk control decision unit, when continuing to train the initialized risk control decision network parameter learning, the target risk control decision unit is trained, and the neuron weight parameters of the first risk control knowledge mining unit, the second risk control knowledge mining unit, and the risk control decision unit among the multiple risk control decision units that meet the training termination requirements are locked.

9. The method for processing risk control information based on Internet of Vehicles according to claim 1, characterized in that: The step of generating the risk control response matching degree of the plurality of reference risk control response strategies for the target Internet of Vehicles user data in the target risk dimension based on the risk prediction data of the target risk dimension corresponding to the risk control management requirement and the obtained priori risk control portrait feature data comprises: Perform feature mapping and correlation analysis on the risk prediction data of the target risk dimension and the prior risk control portrait feature data of each reference risk control response strategy under the target risk dimension from multiple dimensions, and construct a corresponding multi-dimensional correlation analysis result matrix, wherein the multi-dimensional correlation analysis result matrix is ​​used to reflect the relationship between the risk prediction data of the target risk dimension and the prior risk control portrait feature data of each reference risk control response strategy under the target risk dimension, wherein, in the risk factor dimension, various risk factors in the risk prediction data and the response plans for responding to the risk factors in the prior risk control portrait feature data are mapped and analyzed, in the risk trend dimension, the risk trend in the risk prediction data and the response results of the risk trend corresponding to the historical response in the prior risk control portrait feature data are analyzed, and, in the resource utilization dimension, the resource input information represented by the risk prediction data is compared with the resource allocation mode in the prior risk control portrait feature data, and the performance of different reference risk control response strategies in the resource utilization efficiency dimension is evaluated, and, in the execution process dimension, the matching degree between the risk occurrence scenario involved in the risk prediction data and the risk control measure execution process in the prior risk control portrait feature data is analyzed; According to the multi-dimensional correlation analysis result matrix, various evaluation indicators in the quantitative evaluation indicator system are generated, specifically: based on the correlation analysis results of the risk factor dimension, for each risk factor and its corresponding response plan, combined with the importance weight of the risk factor and the ability of the response plan to suppress or mitigate the risk factor, a quantitative risk factor response effectiveness evaluation indicator is constructed, and, using the correlation analysis results of the risk trend dimension, the risk trend adaptability performance of each reference risk control response strategy is quantified, and a rule-based evaluation algorithm is constructed based on the change amplitude and frequency of the risk trend and the adjustment speed and effect of the reference risk control response strategy under different risk trends to generate a quantitative risk trend adaptability evaluation indicator, and, based on the correlation analysis results of the resource utilization dimension, a quantitative resource utilization efficiency evaluation indicator is constructed, and, based on the correlation analysis results of the execution process dimension, an evaluation indicator for the rationality of the quantitative execution process is constructed; A hierarchical structure model is constructed using the analytic hierarchy process, with the Internet of Vehicles environmental factors and risk dimension contribution values ​​as the criterion layer, and each evaluation indicator as the solution layer. The relative importance of the elements in each layer is determined, and the comprehensive weight vector of each evaluation indicator is calculated. According to the implementation of the quantitative evaluation indicator system and the comprehensive weight vector of each evaluation indicator, for each reference risk control response strategy, the value of the reference risk control response strategy on each quantitative evaluation indicator is multiplied by the weight value in the corresponding comprehensive weight vector to obtain the weighted indicator value, and all weighted indicator values ​​are summed to obtain a comprehensive evaluation value. The comprehensive evaluation value serves as the risk control response matching degree of each reference risk control response strategy for the target Internet of Vehicles user data in the target risk dimension, reflecting the matching degree of each reference risk control response strategy with the risk prediction data of the target risk dimension under multiple factors.

10. A risk control information processing system based on Internet of Vehicles, characterized in that: The risk control information processing system based on the Internet of Vehicles includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the risk control information processing method based on the Internet of Vehicles as described in any one of claims 1 to 9.

Images