Risk control big data mining method and system based on Internet of Vehicles

Abstract

The invention provides a risk control big data mining method and system based on the Internet of Vehicles, and the method comprises the steps: collecting a real-time operation data flow of a target vehicle in an Internet of Vehicles environment, including vehicle state time sequence data, a driving behavior event set, external environment perception data and a vehicle-mounted communication interaction record; and then respectively extracting a combined operation feature vector, an abnormal behavior mode identifier and a dynamic environment risk factor from different data, converting to generate a structured communication feature matrix, constructing a space-time association risk analysis map based on the structured communication feature matrix, and carrying out driving risk dynamic assessment on a target vehicle through the space-time association risk analysis map. And finally, generating a risk control decision instruction set according to the result, and feeding back the risk control decision instruction set to the vehicle-mounted control terminal and the Internet of Vehicles cloud platform, thereby realizing comprehensive assessment and effective management and control of vehicle risks.

Description

Technical Field

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

[0002] Against the backdrop of the rapid development of today's intelligent transportation sector, Internet of Vehicles technology has gradually become a hot topic in research and application. It aims to improve traffic efficiency and safety through information interaction between vehicles and the external environment, and to provide drivers with smarter and more convenient services.

[0003] At present, in terms of vehicle risk management in the Internet of Vehicles environment, most existing technologies remain at the simple statistics and classification of some common driving behaviors, and it is difficult to deeply explore potential abnormal behavior patterns. These methods often do not take into account the interrelationships between different driving behaviors and the dynamic changes of behaviors in the time dimension, and cannot timely and accurately identify those abnormal driving behaviors that may cause safety risks. 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 big data mining method based on the Internet of Vehicles, the method comprising: Collecting real-time operation data streams generated by the target vehicle in the Internet of Vehicles environment, the real-time operation data streams including vehicle status time series data, driving behavior event sets, external environment perception data, and vehicle communication interaction records; extracting a combined operation feature vector from the vehicle state time series data, extracting an abnormal behavior pattern identifier from the driving behavior event set, parsing a dynamic environmental risk factor from the external environment perception data, and converting the vehicle communication interaction record into a structured communication feature matrix; Based on the combined operation feature vector, the abnormal behavior pattern identifier, the dynamic environment risk factor and the structured communication feature matrix, construct a spatiotemporal correlation risk analysis map; Performing a dynamic driving risk assessment on the target vehicle through the spatiotemporal correlation risk analysis graph to generate a risk level quantitative index and a risk event warning label; A risk control decision instruction set is generated according to the risk level quantitative index and the risk event warning label, and the risk control decision instruction set is fed back to the on-board control terminal of the target vehicle and the Internet of Vehicles cloud platform.

[0005] On the other hand, an embodiment of the present invention also provides a risk control big data mining 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.

[0006] Based on the above aspects, after collecting the multi-dimensional real-time operation data stream of the target vehicle in the Internet of Vehicles environment, the embodiment of the present application extracts key features from different types of data, including combined operation feature vectors, abnormal behavior pattern identifiers, dynamic environmental risk factors, and structured communication feature matrices, and then constructs a spatiotemporal correlation risk analysis map, thereby deeply integrating and analyzing multi-source heterogeneous data, and accurately capturing the complex spatiotemporal risk relationship during vehicle operation, showing the evolution of vehicle operation risks in time and space dimensions. Furthermore, the constructed spatiotemporal correlation risk analysis map is used to dynamically evaluate the driving risk of the target vehicle, and the risk level quantitative indicators and risk event warning labels can be generated in real time. The risk assessment results can be adjusted in time according to the real-time operation of the vehicle, providing timely and accurate risk information for the driver and the relevant management platform, and effectively responding to the changing risk conditions during vehicle driving. Finally, a risk control decision instruction set is generated based on the risk level quantitative indicators and risk event warning labels, and fed back to the vehicle control terminal and the Internet of Vehicles cloud platform. This enables the Internet of Vehicles system to quickly make targeted decisions based on risk assessment results, achieve proactive intervention and effective control of vehicle risks, ensure safe driving of vehicles, and improve the safety and reliability of the entire Internet of Vehicles ecosystem. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 It is a schematic diagram of the execution flow of the risk control big data mining method based on the Internet of Vehicles provided by an embodiment of the present invention.

[0008] Figure 2 It is a schematic diagram of exemplary hardware and software components of a risk control big data mining system based on the Internet of Vehicles provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 The figure is a flow chart of a risk control big data mining method based on the Internet of Vehicles provided by an embodiment of the present invention. The risk control big data mining method based on the Internet of Vehicles is introduced in detail below.

[0010] Step S110, collecting the real-time operation data stream generated by the target vehicle in the Internet of Vehicles environment, wherein the real-time operation data stream includes vehicle status time series data, driving behavior event set, external environment perception data, and vehicle communication interaction records.

[0011] In detail, taking an electric car driving on a city road as the target vehicle, in the Internet of Vehicles environment, the vehicle's electronic control unit (ECU) continuously records various data to form a real-time operation data stream. In terms of vehicle status time series data, the engine speed sequence records the speed changes of the electric vehicle motor, such as the motor speed gradually increases when the vehicle starts, maintains a stable speed when driving at a constant speed, and rapidly decreases when braking. The brake pressure waveform reflects the pressure changes during each braking operation, the battery temperature curve changes with factors such as vehicle driving time, charging and discharging cycles, and ambient temperature, and the tire wear increment gradually accumulates with the increase in mileage. Driving behavior events are concentrated, recording the driver's operating events, such as the driver's sudden acceleration when encountering a sudden deceleration of the vehicle in front, frequent continuous lane changes on multi-lane roads, speeding on certain sections of the road, and not using lights as required at night. In terms of external environment perception data, the sensors equipped on the vehicle perceive real-time meteorological condition parameters, such as raindrop sensors detecting rainfall intensity and reduced visibility on rainy days; the camera identifies the heat map of road obstacle distribution, such as a vehicle parked on the road due to a malfunction, which is displayed as a high-risk obstacle gathering area on the heat map; at the same time, it can also accurately obtain traffic sign semantic information, such as recognizing that the speed limit sign ahead requires the speed not to exceed 60 kilometers per hour, and the movement trajectory set of surrounding vehicles, such as the driving direction, speed and distance between the surrounding vehicles and the target vehicle. The vehicle-mounted communication interaction record contains the communication data between the vehicle and other nearby vehicles and roadside infrastructure (such as smart traffic lights and roadside units), such as the vehicle communicating with the traffic light to obtain information such as the remaining time of the light.

[0012] Step S120, extracting a combined operation feature vector from the vehicle state time series data, extracting an abnormal behavior pattern identifier from the driving behavior event set, parsing dynamic environmental risk factors from the external environment perception data, and converting the vehicle communication interaction record into a structured communication feature matrix.

[0013] In detail, for the engine speed sequence in the vehicle state time series data, when the frequency domain transformation is performed, for example, the periodic fluctuation amplitude characteristics of the motor speed of an electric vehicle during acceleration can reflect the working stability of the motor. If the speed mutation detection mark shows a sudden speed drop during normal driving, it may indicate that the motor or transmission system is faulty. When the brake pressure waveform is processed by segmented integration to generate a brake intensity distribution histogram, if it is found that the intensity of most braking operations is concentrated in a lower area, but occasionally there is an abnormal pressure fluctuation threshold of high-intensity braking, this may indicate that the driver's driving habits or the brake system are abnormal. After the trend decomposition of the battery temperature curve, the long-term temperature rise gradient characteristics show that the battery temperature continues to rise with a large slope during long-term driving, and the short-term temperature oscillation parameters show that the battery temperature has a rapid small fluctuation under certain working conditions, which may affect the battery life and performance. After the tire wear increment is modeled by the wear rate, the calculated tire life prediction index shows that the remaining life of the tire is short and the probability of abnormal tread loss is high, which may be due to the vehicle's long-term driving on roads with poor road conditions or bad driving habits. Therefore, the above features can be normalized and spliced ​​to form a combined operation feature vector.

[0014] When extracting abnormal behavior pattern identifiers from the driving behavior event set, the number of times the sudden acceleration event is triggered is screened out. For example, if the sudden acceleration event is triggered 5 times within an hour, a higher sudden acceleration risk coefficient is generated after comparison with the preset acceleration threshold. The statistics of the time intervals of continuous lane change events show that the time interval between drivers changing lanes continuously on a section of road is very short, which determines a higher lane change behavior density score. The speeding duration ratio calculation shows that the time of speeding in a section of road accounts for 30% of the total driving time, and a large speeding violation accumulation is generated in combination with the road speed limit data. The frequency of illegal lighting use at night is high, and the lighting violation pattern identifier is generated by matching with the lighting specification database. These data are weighted and fused to generate an abnormal behavior pattern identifier.

[0015] The dynamic environmental risk factors are parsed from the external environment perception data, and the real-time meteorological condition parameters are identified as rainy weather. The visibility attenuation coefficient is calculated to be high, and the road adhesion risk index is also increased due to rain. The road obstacle distribution heat map shows that there is a faulty vehicle on the right side of the road ahead. The obstacle collision probability distribution map generated based on its density gradient shows that the target vehicle has a high collision probability when it approaches the faulty vehicle. The traffic sign semantic information is checked for compliance. If it is found that the vehicle is about to exceed the speed limit, the sign violation risk weight is generated. The relative speed cluster analysis is performed through the surrounding vehicle motion trajectory set. For example, when it is found that the surrounding vehicles are fast and the distance from the target vehicle is small, the vehicle spacing safety assessment matrix is ​​generated. Then, these visibility attenuation coefficients, road adhesion risk indexes, obstacle collision probability distribution maps, sign violation risk weights and vehicle spacing safety assessment matrices are spatially superimposed. With the target vehicle as the center, the spatial area is divided into square grid units according to a certain radius. Various risk factors are calculated and superimposed in each grid unit to finally generate dynamic environmental risk factors.

[0016] When converting the vehicle-mounted communication interaction records into a structured communication feature matrix, such as the communication interaction between a vehicle and multiple roadside units, the time, frequency, data volume and other information of the communication are recorded, and this information is organized into a matrix form according to certain rules to reflect the communication status and relationship between the vehicle and external devices.

[0017] Step S130, constructing a spatiotemporal correlation risk analysis graph based on the combined operation feature vector, the abnormal behavior pattern identifier, the dynamic environment risk factor and the structured communication feature matrix.

[0018] In detail, the combined operation feature vector can be mapped to the state attribute node of the target vehicle, such as mapping the engine speed, brake pressure and other related features to the node representing the vehicle state. The abnormal behavior pattern identifier is mapped to the driving behavior associated edge. For example, the situation with a high risk coefficient of sudden acceleration is mapped to an edge indicating that the risk of sudden acceleration affects the vehicle state. The dynamic environmental risk factor is mapped to the environmental constraint layer. Factors such as reduced visibility caused by rainy weather and increased road adhesion risk constitute environmental constraints on vehicle driving. The structured communication feature matrix is ​​mapped to the communication interaction path, such as the communication path between the vehicle and the roadside unit.

[0019] The state attribute nodes are divided into time slices based on the preset time window, for example, 5 minutes as a time slice, and the changes in the vehicle state are analyzed in each slice to generate the time dimension topological relationship. The environmental constraint layer is spatially gridded according to the vehicle's geographic location data. Assuming that the vehicle is the center, the surrounding 100 meters is divided into several square spatial grid units to generate a spatial dimension association matrix. The state attribute nodes in the same time slice are connected through driving behavior association edges. For example, the rapid acceleration behavior association edge connects the motor speed change node and the brake pressure change node in a time slice to form a driving behavior impact link. The communication interaction path is cross-layer coupled with the driving behavior impact link. For example, after the vehicle communicates with the roadside unit to obtain traffic information, it affects the driver's driving behavior and generates a communication delay risk propagation channel.

[0020] When integrating the above elements, the time stamp alignment processing is performed on each time slice division result in the time dimension topological relationship, so that the time order of each time slice is accurate and a time topological node chain with continuous time series attributes is generated. Each spatial grid unit in the spatial dimension association matrix is ​​bound to the corresponding time slice in the time topological node chain by time and space coordinates, for example, a grid unit is associated with the corresponding vehicle state at a certain moment to generate a time and space coding grid unit. The starting state attribute node and the ending state attribute node of each driving behavior association edge in the driving behavior influence link are extracted, and mapped to the time slice index and space grid coordinates of the time and space coding grid unit. According to the time slice index of the time and space coding grid unit, the continuous multiple driving behavior association edges in the driving behavior influence link are spliced ​​by time window sliding to generate the driving behavior propagation path across time slices. The transmission delay parameter of each communication interaction path is extracted from the communication delay risk propagation channel, and the driving behavior propagation path across time slices is offset by delay compensation according to the parameter to generate the driving behavior propagation path after delay calibration. Each node in the delay-calibrated driving behavior propagation path is matched with the spatial grid coordinates of the spatiotemporal coding grid unit to generate a driving behavior propagation path with spatial position constraints. The dynamic environmental risk factor of the environmental constraint layer is superimposed on the spatial grid coordinates of the spatiotemporal coding grid unit to generate the risk density weight of each spatiotemporal coding grid unit. The driving behavior propagation path with spatial position constraints is weighted according to the risk density weight to generate a risk transmission probability distribution path. Each node in the risk transmission probability distribution path is gradient-connected with the risk density weight of the adjacent spatiotemporal coding grid unit to generate a risk gradient transmission link. Time slice back-testing is performed on each node in the risk gradient transmission link to identify the node set whose risk density weight exceeds the threshold in continuous time slices and generate a risk spatiotemporal focus area. According to the spatial grid coordinate distribution of the risk spatiotemporal focus area, multiple nodes in the risk gradient transmission link located at the same spatial grid coordinate are path-aggregated to generate a spatially aggregated risk transmission link. Each path node in the spatially aggregated risk transmission link is time-dimension-expanded, and the path nodes of the same spatial grid coordinate at different time slices are connected as a time-evolved risk transmission trajectory. The time-evolved risk transmission trajectory is cross-fused with the risk gradient transmission link to generate a multi-level risk transmission network with spatiotemporal evolution attributes. Each node in the multi-level risk transmission network is dynamically mapped to the risk density weight of the spatiotemporal coding grid unit to generate a node risk weight label. The transmission paths in the multi-level risk transmission network are ranked according to the node risk weight labels to generate a priority-labeled spatiotemporal correlation risk transmission map.Finally, the priority marks in the spatiotemporal correlation risk transmission map are spatially superimposed and verified with the dynamic environmental risk factors of the environmental constraint layer to eliminate the transmission paths with spatial grid coordinate conflicts. Based on the verified spatiotemporal correlation risk transmission map and the time slicing results of the spatiotemporal coding grid units, a spatiotemporal correlation risk analysis map with spatiotemporal stamps is generated.

[0021] Step S140, dynamically assessing the driving risk of the target vehicle through the spatiotemporal correlation risk analysis graph, and generating a risk level quantitative index and a risk event warning label.

[0022] In detail, the driving behavior impact link is traversed in the spatiotemporal correlation risk analysis map, and the risk transmission intensity of each link is calculated. For example, in a driving behavior impact link, the risk transmission intensity of a series of impacts such as sudden changes in motor speed and increased brake pressure caused by sudden acceleration behavior is relatively high. According to the path length and delay time of the communication delay risk propagation channel, the risk response hysteresis coefficient is determined. For example, when the communication delay between the vehicle and the roadside unit is long, the risk response hysteresis coefficient is large. Based on the spatial grid density and risk factor weights of the environmental constraint layer, the environmental risk superposition effect value is generated. In areas with rainy weather and obstacles on the road, the environmental risk superposition effect value is higher. The characteristic vector of the state attribute node is monitored in real time to generate a vehicle state degradation index. For example, the battery temperature rises too fast, the tire wear rate suddenly increases, etc., which will increase the vehicle state degradation index.

[0023] The risk transmission intensity, risk response hysteresis coefficient, environmental risk superposition effect value and vehicle state degradation index are input into the risk quantification model to output the risk level quantitative index. Assuming that the risk level quantitative index is 80 (out of 100), it means that the risk is relatively high. According to the comparison between the risk level quantitative index and the preset threshold, different levels of risk event warning labels are triggered. When the risk level quantitative index reaches 80, a high-level risk event warning label is triggered, such as "High Risk-Urgent Attention", indicating that the vehicle may face a more serious risk situation and timely measures need to be taken.

[0024] Step S150, generating a risk control decision instruction set according to the risk level quantitative index and the risk event warning label, and feeding back the risk control decision instruction set to the on-board control terminal of the target vehicle and the Internet of Vehicles cloud platform.

[0025] In detail, the basic risk control strategy type can be determined according to the risk level quantitative index. When the risk level quantitative index is 80, it is determined to be an emergency braking instruction. Combined with the label type "high risk-urgent concern" of the risk event warning label, additional control parameters are matched from the vehicle network policy library, such as the parameters for turning on the hazard warning lights at the same time during emergency braking. Based on the real-time location data of the target vehicle and the high-precision map information, the strategy effective range under the geographical constraints is generated. For example, the topological structure data of the road where the target vehicle is currently located shows that there are many curves, the real-time traffic flow information shows that the vehicle density is large, and the high-precision map information shows that the road curvature radius is small, the slope is steep, and it is a two-way two-lane road. Combined with the risk level quantitative index, the curve control radius is generated, such as further reducing the safe passing speed of the curve; according to the vehicle density distribution and the risk event warning label, the avoidance area priority is generated, such as giving priority to avoiding vehicles in the right lane. Integrating this information defines the spatial boundary and time validity period of the strategy effective range, such as within the road range of 200 meters ahead and effective within the next 10 minutes.

[0026] Through the hardware interface status detection of the vehicle control terminal, the executable instruction set is determined and the equipment conflict instructions are excluded. For example, if the braking system is detected to be normal, the emergency braking instruction can be executed, while some instructions that conflict with the current operation are excluded. The basic risk control strategy type, additional control parameters, strategy effective range and executable instruction set are logically combined to generate a risk control decision instruction set, and then the instruction set is fed back to the vehicle control terminal of the target vehicle and the Internet of Vehicles cloud platform. After receiving the instruction set, the Internet of Vehicles cloud platform can comprehensively manage the risk situation of multiple vehicles, such as conducting a global risk situation analysis, generating a regional risk heat map based on the geographical distribution of multiple target vehicles, adjusting the traffic signal control strategy and the roadside unit broadcast content based on the regional risk heat map, sending the adjusted traffic signal control strategy to the intelligent traffic management system, and updating the roadside unit broadcast content. According to the aggregation results of the risk level quantitative indicators of multiple target vehicles, a macro Internet of Vehicles risk trend report is generated. For example, if the quantitative indicators of risk levels of multiple vehicles in a certain area are high, it will be displayed as a high-risk area on the regional risk heat map. The traffic signal control strategy in the area will be adjusted to extend the red light time to reduce vehicle flow, and the roadside unit broadcast content will be updated to remind drivers to pay attention to the high-risk area ahead and drive carefully.

[0027] Based on the above steps, after the embodiment of the present application collects the multi-dimensional real-time operation data stream of the target vehicle in the Internet of Vehicles environment, key features are extracted from different types of data, including combined operation feature vectors, abnormal behavior pattern identifiers, dynamic environmental risk factors, and structured communication feature matrices, and then a spatiotemporal correlation risk analysis map is constructed, so that multi-source heterogeneous data can be deeply integrated and correlated, and the complex spatiotemporal risk relationship in the vehicle operation process can be accurately captured, showing the evolution law of vehicle operation risk in time and space dimensions. Furthermore, the target vehicle is dynamically evaluated for driving risk using the constructed spatiotemporal correlation risk analysis map, which can generate risk level quantitative indicators and risk event warning labels in real time, and can adjust the risk assessment results in time according to the real-time operation of the vehicle, provide timely and accurate risk information for drivers and related management platforms, and effectively respond to the changing risk conditions during vehicle driving. Finally, a risk control decision instruction set is generated based on the risk level quantitative indicators and risk event warning labels, and fed back to the vehicle control terminal and the Internet of Vehicles cloud platform. This enables the Internet of Vehicles system to quickly make targeted decisions based on risk assessment results, achieve proactive intervention and effective control of vehicle risks, ensure safe driving of vehicles, and improve the safety and reliability of the entire Internet of Vehicles ecosystem.

[0028] In a possible implementation manner, the vehicle state time series data includes an engine speed sequence, a brake pressure waveform, a battery temperature curve, and a tire wear increment, and step S120 includes: Step S1201, performing frequency domain transformation on the engine speed sequence to extract periodic fluctuation amplitude characteristics and speed mutation detection marks.

[0029] In this embodiment, for the engine speed sequence in the vehicle state time series data, the speed of the motor is constantly changing during the operation of the electric vehicle. When the engine speed sequence is transformed in the frequency domain, it can be understood as analyzing the speed data from another perspective. For example, during the normal acceleration of the vehicle, the motor speed will rise regularly, and the periodic fluctuation amplitude characteristics can be extracted through the frequency domain transformation in this process. Assuming that this amplitude feature is in a stable range under normal circumstances, this reflects the stable operation state of the motor under normal working conditions. The speed mutation detection mark is used to detect whether there is an abnormal sudden change in speed during normal driving. For example, when the vehicle is driving on a flat road, the motor speed suddenly drops sharply for an instant, and this situation will be recorded by the speed mutation detection mark, which may indicate that there is a potential failure risk in the motor or transmission system.

[0030] Step S1202, performing segmented integration processing on the brake pressure waveform to generate a brake intensity distribution histogram and a pressure abnormal fluctuation threshold.

[0031] In this embodiment, in terms of the brake pressure waveform, each time the driver steps on the brake pedal, the brake system will generate a corresponding change in brake pressure. The brake pressure waveform is processed by segmented integration, and this process is like a detailed analysis of the entire braking process according to different stages. For example, in a long braking process, a brake intensity distribution histogram can be generated through segmented integration. Assume that this histogram shows that most of the braking intensity is concentrated in a relatively low area, which means that the driver's braking operation is relatively gentle in most cases. However, if there are some special intervals where the braking intensity suddenly becomes very high, this situation will generate an abnormal pressure fluctuation threshold. For example, during emergency braking, the brake pressure will rise sharply. The appearance of this abnormal pressure fluctuation threshold can reflect this special braking situation, and may also indicate that there is something abnormal in the braking system or driving habits.

[0032] Step S1203 , performing trend decomposition on the battery temperature curve to separate long-term temperature rise gradient characteristics and short-term temperature oscillation parameters.

[0033] In this embodiment, the battery temperature curve reflects the temperature change of the battery during the driving process of the vehicle. By performing trend decomposition on the battery temperature curve, the inherent law of battery temperature change can be more clearly understood. During long-term driving, the battery temperature will gradually rise, and the long-term temperature rise gradient characteristics can be separated through trend decomposition. Suppose that during a long-distance driving, it is found that the long-term temperature rise gradient characteristics of the battery temperature show a relatively steep upward trend, which may mean that there is a problem with the battery's heat dissipation system or that the performance of the battery itself is gradually declining. At the same time, during the driving of the vehicle, due to some special working conditions, such as frequent acceleration and deceleration, the battery temperature will have some small rapid fluctuations, and these fluctuations can be decomposed to obtain short-term temperature oscillation parameters. For example, when the vehicle frequently starts and stops on urban roads, the battery temperature will have some short-term ups and downs, and the fluctuation is recorded by the short-term temperature oscillation parameters.

[0034] Step S1204, performing wear rate modeling on the tire wear increment, and calculating the tire life prediction index and the probability of abnormal tread wear.

[0035] In this embodiment, the tire wear increment gradually accumulates as the vehicle mileage increases. The wear rate modeling of the tire wear increment can be understood as building a mathematical model to accurately describe the speed of tire wear. The calculated tire life prediction index can be used to estimate how long the tire can be used. For example, if the tire life prediction index shows that the tire has only 20% of its remaining life, this prompts the driver to pay attention to the condition of the tire and consider whether the tire needs to be replaced. The probability of abnormal tread wear reflects the possibility of abnormal tire wear. Assuming that the vehicle often travels on roads with poor road conditions, such as roads full of potholes and stones, the probability of abnormal tread wear will be relatively high, which may be due to the uneven road surface causing increased tire wear.

[0036] Step S1205, normalize and splice the periodic fluctuation amplitude characteristics, the speed mutation detection mark, the braking intensity distribution histogram, the pressure abnormal fluctuation threshold, the long-term temperature rise gradient characteristics, the short-term temperature oscillation parameters, the tire life prediction index and the tread abnormal wear probability to form the combined operation feature vector.

[0037] In this embodiment, the combined operation characteristic vector is a comprehensive index set that comprehensively reflects various state characteristics of the vehicle during operation. For example, the values ​​of each element in the combined operation characteristic vector can be input into a subsequent analysis system to evaluate the overall operation health of the vehicle.

[0038] In a possible implementation, step S120 further includes: Step S1206, filtering the number of sudden acceleration event triggering times, the time interval of continuous lane change events, the proportion of speeding duration, and the frequency of illegal light use at night from the driving behavior event set.

[0039] Step S1207: Generate a sudden acceleration risk coefficient based on the comparison between the sudden acceleration event triggering times and a preset acceleration threshold.

[0040] For example, during the driving of this electric car, the number of times the sudden acceleration event is triggered is an important indicator. Assume that in a specific driving distance, the sudden acceleration event is triggered 3 times. By comparing the number of sudden acceleration event triggers with the preset acceleration threshold, a sudden acceleration risk factor can be generated. If the preset acceleration threshold is a certain value, and the acceleration in these 3 sudden acceleration events exceeds this threshold, then the sudden acceleration risk factor will be relatively high, indicating that the driver's sudden acceleration operation may pose a greater risk to the safety and performance of the vehicle.

[0041] Step S1208: determining a lane changing behavior density score according to the statistical distribution of the time intervals between the consecutive lane changing events.

[0042] On multi-lane urban roads, drivers may change lanes continuously. By counting the time intervals between consecutive lane change events, the lane change behavior density score can be determined. Assuming that on a road, a driver changes lanes three times in a short distance, and the time interval between each lane change is very short, this will result in a high lane change behavior density score. This means that the driver's lane change behavior is relatively frequent and hasty, increasing the risk of collision with other vehicles.

[0043] Step S1209, generating a speeding violation accumulation amount by combining the speeding duration ratio and the road speed limit data.

[0044] During the entire driving process, the speed limit data is combined to calculate the proportion of speeding duration. For example, on a road with a speed limit of 60 km / h, the total driving time of the vehicle is 30 minutes, of which the speeding time (speed exceeding 60 km / h) is 10 minutes, then the speeding duration ratio is 10 / 30≈33.3%. This speeding duration ratio is combined with the road speed limit data to generate the speeding violation accumulation. If the speeding violation accumulation is large, it means that the driver has a more serious speeding violation on this road, increasing the risk of traffic accidents.

[0045] Step S1210, generating a light violation pattern identifier by matching the nighttime illegal light usage frequency with a light specification database.

[0046] When driving at night, there are clear regulations for the use of vehicle lights. By counting the frequency of nighttime illegal light use and then matching it with the light specification database, a light violation pattern mark can be generated. For example, if a driver often fails to use high beams or turn signals at night as required, the frequency of nighttime illegal light use will be high, and the generated light violation pattern mark will show this illegal behavior pattern.

[0047] Step S1211, weighted fusion of the sudden acceleration risk coefficient, the lane change behavior density score, the speeding violation accumulation amount and the light violation pattern identifier to generate the abnormal behavior pattern identifier.

[0048] In this embodiment, the abnormal behavior pattern identifier can be understood as a comprehensive evaluation result, which comprehensively reflects whether the driver's driving behavior has abnormal risks. For example, if the risk coefficient of sudden acceleration is high, the lane change behavior density score is high, the accumulated speeding violations are large, and the light violation pattern identifier shows more violations, then the weighted fusion abnormal behavior pattern identifier will show a higher risk level, indicating that the driver's driving behavior needs to be corrected or monitored to ensure the safety of the vehicle.

[0049] In a possible implementation, step S120 further includes: Step S1212, identifying real-time meteorological condition parameters, road obstacle distribution heat map, traffic sign semantic information and surrounding vehicle motion trajectory set from the external environment perception data.

[0050] Step S1213, calculating the visibility attenuation coefficient and the road adhesion risk index according to the real-time meteorological condition parameters.

[0051] Step S1214: generating an obstacle collision probability distribution map based on the density gradient of the road obstacle distribution heat map.

[0052] Step S1215: Perform compliance check on the traffic sign semantic information and generate a sign violation risk weight.

[0053] Step S1216: performing relative speed cluster analysis on the surrounding vehicle motion trajectory set to generate a vehicle spacing safety assessment matrix.

[0054] For real-time meteorological condition parameters, sensors will continuously monitor meteorological conditions during the driving of electric vehicles. For example, when encountering rainy weather, meteorological sensors will detect parameters such as rainfall intensity and humidity, and calculate the visibility attenuation coefficient and road adhesion risk index based on these real-time meteorological condition parameters. Rainfall will cause visibility to decrease, and the visibility attenuation coefficient is calculated through an algorithm. Assuming that in a moderate rainy day, visibility decreases from the normal several thousand meters to several hundred meters, the proportion and degree of this decrease is reflected in the visibility attenuation coefficient. At the same time, rainfall will make the road surface slippery, and different road surface materials have different adhesion capabilities in a wet state. The road adhesion risk index is calculated based on this situation. For example, for asphalt pavement, the road adhesion risk index will increase significantly after rainfall, indicating that the vehicle is more likely to slip when braking and turning.

[0055] On urban roads, there are many other vehicles around, and the vehicle's sensors track the movement trajectories of these surrounding vehicles. By performing relative speed cluster analysis on the surrounding vehicle movement trajectory set, for example, vehicles with similar speeds and relatively stable spacing are classified into one category. If it is found that the speed of the surrounding vehicles is fast and the distance to the target electric vehicle is small, the corresponding evaluation value will be generated in the vehicle spacing safety assessment matrix. Assuming that on a multi-lane road, the vehicles in the adjacent lanes are approaching the target vehicle quickly and the distance is constantly decreasing, the evaluation value in the vehicle spacing safety assessment matrix will show a higher risk level.

[0056] Step S1217, spatially superimposing the visibility attenuation coefficient, the road adhesion risk index, the obstacle collision probability distribution map, the sign violation risk weight and the vehicle spacing safety assessment matrix to generate the dynamic environment risk factor.

[0057] For example, step S1217 includes: Step S1217-1, based on the geographic coordinate information in the real-time meteorological condition parameters and the coverage of the road obstacle distribution heat map, the spatial area within a preset radius around the target vehicle is divided into square grid units of equal size to generate an environmental space overlay grid.

[0058] For example, with the target electric car as the center, the area with a radius of 500 meters is divided into square grid units with a side length of 10 meters.

[0059] Step S1217-2, performing spatial interpolation calculation on the visibility attenuation coefficient according to the center position of each grid cell of the environmental space superposition grid, and generating a visibility attenuation coefficient distribution grid corresponding to each grid cell.

[0060] Assuming that at the center of a grid cell, the visibility attenuation coefficient distribution grid value of the grid cell is obtained by spatial interpolation calculation based on the distance from the surrounding meteorological sensor detection points and the relationship between the visibility attenuation coefficient. If the grid cell is close to the rainfall center area, the visibility attenuation coefficient distribution grid value will be higher.

[0061] Step S1217-3, according to the preset mapping relationship of the road adhesion risk index in different road material areas, the road adhesion risk index is matched and filled according to the road material type to which the grid unit of the environmental space superimposed grid belongs, to generate a road adhesion risk index distribution grid.

[0062] For example, the road area covered by a certain grid unit is a cement road surface. According to the preset value of the road adhesion risk index of the cement road surface after rainfall, this value is filled into the road adhesion risk index distribution grid of the grid unit.

[0063] Step S1217 - 4 , performing probability density resampling on the obstacle collision probability distribution map according to the grid unit boundaries of the environmental space superimposed grid to generate a normalized grid of obstacle collision probability for each grid unit.

[0064] If there is a part of the area in a certain grid cell that is close to the faulty vehicle on the road, the normalized grid value of the obstacle collision probability of the grid cell will be higher through probability density resampling, reflecting that the possibility of collision between the vehicle and the obstacle is higher in this grid cell.

[0065] Step S1217-5, based on the geographic location data of the traffic sign semantic information, project the sign violation risk weight into the corresponding grid unit of the environmental space overlay grid to generate a sign violation risk weight distribution grid.

[0066] For example, in a grid cell that is close to a speed limit sign, since there is a risk of speeding for the vehicle, the grid value of the sign violation risk weight distribution will be higher after the sign violation risk weight is projected onto this grid cell.

[0067] Step S1217-6, extracting the relative position coordinates of each surrounding vehicle relative to the target vehicle in the vehicle spacing safety assessment matrix, performing nearest neighbor allocation on the assessment values ​​of the vehicle spacing safety assessment matrix according to the grid cells of the environmental space overlay grid, and generating a vehicle spacing safety assessment distribution grid.

[0068] If the distance between a surrounding vehicle and the target vehicle is small and is located near a grid cell, the vehicle distance safety assessment distribution grid value of the grid cell will reflect the risk condition of this close distance.

[0069] Step S1217-7, uniformly normalize the visibility attenuation coefficient distribution grid, the road adhesion risk index distribution grid, the obstacle collision probability normalization grid, the sign violation risk weight distribution grid and the vehicle spacing safety assessment distribution grid so that the value range of each grid is mapped to the same risk level interval.

[0070] For example, the values ​​of the visibility attenuation coefficient distribution grid are converted from the original actual measurement range to a value between 0 and 1, and other grids are also converted similarly to ensure that they are under the same risk level evaluation system.

[0071] Step S1217-8, based on the preset visibility risk weight, road adhesion risk weight, obstacle collision risk weight, sign violation risk weight and vehicle spacing risk weight, the normalized distribution grids are weighted and summed according to the corresponding grid cells to generate a comprehensive risk superposition value for each grid cell.

[0072] Assuming that the visibility risk weight is 0.2, the road adhesion risk weight is 0.3, the obstacle collision risk weight is 0.2, the sign violation risk weight is 0.1, and the vehicle spacing risk weight is 0.2, in a certain grid unit, the comprehensive risk superposition value is obtained through weighted summation calculation.

[0073] Step S1217-9, based on the distribution of the comprehensive risk superposition value, spatial smoothing filtering is performed on the comprehensive risk superposition value of each grid cell in the environmental space superposition grid to eliminate noise interference of isolated high-risk grid cells.

[0074] For example, a grid cell may have an extremely high comprehensive risk superposition value due to a short-term error in the sensor, but the risk values ​​of the surrounding grid cells are lower. Spatial smoothing filtering can reduce the risk value of this isolated high-risk grid cell, making it more consistent with the actual risk distribution.

[0075] Step S1217-10, segment and mark the comprehensive risk superposition value after spatial smoothing filtering according to the preset risk level division threshold, and generate a dynamic environmental risk factor spatial distribution map containing multi-level risk areas.

[0076] For example, the comprehensive risk superposition value is divided into three levels: low risk (0-0.3), medium risk (0.3-0.6) and high risk (0.6-1). In the dynamic environmental risk factor spatial distribution map, areas of different levels are marked with different colors or symbols.

[0077] Step S1217-11, based on the real-time movement direction and speed vector of the target vehicle, risk propagation rate compensation is performed on the grid cells located within a preset distance in front of the vehicle in the dynamic environmental risk factor spatial distribution map to generate a dynamic environmental risk factor with an expanded time dimension.

[0078] Assuming that the target electric vehicle is moving forward at a certain speed, for the grid cells within 200 meters in front of the vehicle, considering that the vehicle will enter these areas in the future, the risk values ​​of these grid cells are adjusted according to the vehicle's speed and direction of travel to reflect the spread of risk in the time dimension. For example, if there is a high-risk area in front of the vehicle and the vehicle is approaching quickly, then after the risk propagation rate compensation, the risk value of the high-risk area will be further increased, thereby more accurately reflecting the risk situation that the vehicle will face.

[0079] In a possible implementation, step S130 includes: Step S131, mapping the combined operation feature vector to the state attribute node of the target vehicle, mapping the abnormal behavior pattern identifier to the driving behavior association edge, mapping the dynamic environment risk factor to the environment constraint layer, and mapping the structured communication feature matrix to the communication interaction path.

[0080] First, the combined operation feature vector is mapped to the state attribute node of the target vehicle. For example, the elements in the combined operation feature vector, such as the periodic fluctuation amplitude characteristics in the engine speed sequence and the speed mutation detection mark, can be mapped to nodes representing the vehicle state. These nodes reflect various aspects of the internal operation state of the vehicle. The abnormal behavior pattern identifier is mapped to the driving behavior association edge. The abnormal behavior pattern identifier composed of the sudden acceleration risk coefficient and the lane change behavior density score can construct the edge representing the driving behavior association, such as the association edge between the sudden acceleration behavior and the vehicle state change. The dynamic environmental risk factor is mapped to the environmental constraint layer. For example, the dynamic environmental risk factor composed of the visibility attenuation coefficient and the road adhesion risk index constitutes the environmental constraint on vehicle driving. For example, the reduced visibility and slippery road surface caused by rain are an environmental constraint. The structured communication feature matrix is ​​mapped to the communication interaction path. For example, the structured communication feature matrix composed of the communication interaction records between the vehicle and the roadside unit or other vehicles reflects the communication path relationship between the vehicle and the outside.

[0081] Step S132: time slice the state attribute node based on a preset time window to generate a time dimension topological relationship.

[0082] Assuming that the preset time window is 5 minutes, the vehicle status attribute nodes are analyzed within every 5-minute time period. For example, the status attribute nodes such as engine speed and brake pressure within 0-5 minutes form a time slice, and the status attribute nodes within 5-10 minutes form another time slice. The order and relationship between these time slices constitute the time dimension topological relationship.

[0083] Step S133, dividing the environmental constraint layer into spatial grids according to the vehicle geographic location data to generate a spatial dimension association matrix.

[0084] For example, with the current position of the vehicle as the center, the surrounding area is divided into square spatial grid units. For example, the 100m×100m area around the vehicle is divided into square grid units with a side length of 10 meters. Each grid unit corresponds to different environmental constraint information, and the relationship between these grid units forms a spatial dimension association matrix.

[0085] Step S134: connecting the state attribute nodes in the same time slice through the driving behavior association edge to form a driving behavior influence link.

[0086] For example, within the same 5-minute time slice, the sudden acceleration behavior association edge may connect the state attribute node of the engine speed mutation and the state attribute node of the brake pressure change, thus constructing a driving behavior influence link, indicating the impact path of driving behavior on the vehicle state during this time period.

[0087] Step S135 , cross-layer coupling is performed between the communication interaction path and the driving behavior impact link to generate a communication delay risk propagation channel.

[0088] For example, a vehicle communicates with a roadside unit to obtain traffic congestion information. The communication interaction path is coupled with a driving behavior influence link in which the driver changes his driving behavior (such as slowing down or changing lanes) due to traffic congestion. Since there may be delays in communication, the delay will affect the adjustment of driving behavior, thereby forming a communication delay risk propagation channel.

[0089] Step S136, integrating the time dimension topological relationship, the space dimension association matrix, the driving behavior impact link and the communication delay risk propagation channel to form the space-time association risk analysis map.

[0090] For example, step S136 includes: Step S1361, performing timestamp alignment processing on each time slice division result in the time dimension topological relationship to generate a time topological node chain with continuous time series attributes.

[0091] Step S1362, binding the space-time coordinates of each space grid unit in the space dimension association matrix to the corresponding time slice in the time topology node chain to generate a space-time coding grid unit.

[0092] Step S1363: extracting the start state attribute node and the end state attribute node of each driving behavior associated edge in the driving behavior influence link, and mapping the start state attribute node and the end state attribute node to the time slice index and the spatial grid coordinate of the spatiotemporal coding grid unit.

[0093] Step S1364: according to the time slice index of the spatiotemporal coding grid unit, a plurality of continuous driving behavior associated edges in the driving behavior influence link are spliced ​​by time window sliding, so as to generate a driving behavior propagation path across time slices.

[0094] Step S1365: extract the transmission delay parameters of each communication interaction path from the communication delay risk propagation channel, and perform delay compensation offset on the driving behavior propagation path across time slices according to the transmission delay parameters to generate a driving behavior propagation path after delay calibration.

[0095] Step S1366, position matching is performed on each node in the driving behavior propagation path after the delay calibration with the spatial grid coordinates of the spatiotemporal coding grid unit to generate a driving behavior propagation path with spatial position constraints.

[0096] Step S1367, superimposing the dynamic environmental risk factor of the environmental constraint layer on the spatial grid coordinates of the spatiotemporal coding grid unit to generate a risk density weight for each spatiotemporal coding grid unit.

[0097] Step S1368, weighting the risk transmission probability of the driving behavior propagation path with spatial position constraints according to the risk density weight, and generating a risk transmission probability distribution path.

[0098] Step S1369, gradient connection is performed between each node in the risk conduction probability distribution path and the risk density weight of the adjacent spatiotemporal coding grid unit to generate a risk gradient conduction link.

[0099] Step S13610, perform time slice backtracking detection on each node in the risk gradient transmission link, identify the node set whose risk density weight exceeds the threshold in continuous time slices, and generate a risk spatiotemporal focus area.

[0100] Step S13611, based on the spatial grid coordinate distribution of the risk spatiotemporal focus area, multiple nodes located at the same spatial grid coordinate in the risk gradient transmission link are path aggregated to generate a spatially aggregated risk transmission link.

[0101] Step S13612, expand the time dimension of each path node in the spatially aggregated risk transmission link, and connect the path nodes of the same spatial grid coordinates in different time slices into a time-evolved risk transmission trajectory.

[0102] Step S13613, cross-integrate the time-evolved risk conduction trajectory with the risk gradient conduction link to generate a multi-level risk conduction network with spatiotemporal evolution properties.

[0103] Step S13614, dynamically map each node in the multi-level risk conduction network with the risk density weight of the spatiotemporal coding grid unit to generate a node risk weight label.

[0104] Step S13615, sorting the risk intensity of the conduction paths in the multi-level risk conduction network according to the node risk weight labels, and generating a priority-marked spatiotemporal correlation risk conduction map.

[0105] Step S13616, spatially superimpose and verify the priority marks in the spatiotemporal correlation risk transmission map with the dynamic environmental risk factors of the environmental constraint layer to eliminate the transmission paths with spatial grid coordinate conflicts, and generate the spatiotemporal correlation risk analysis map with spatiotemporal stamps based on the verified spatiotemporal correlation risk transmission map and the time slicing results of the spatiotemporal coding grid units.

[0106] In this embodiment, the time stamp alignment processing is performed on each time slice division result in the time dimension topological relationship to generate a time topological node chain with continuous time series attributes. Ensure that the start and end time of each time slice are accurately marked, so that each time slice is connected in sequence into a continuous time topological node chain, such as 0-5 minutes, 5-10 minutes and other time slices are accurately connected. Bind the space-time coordinates of each spatial grid unit in the spatial dimension association matrix to the corresponding time slice in the time topological node chain to generate a space-time coding grid unit. For example, in a specific 5-minute time slice, the spatial grid unit at a specific position around the vehicle is bound to it to form a space-time coding grid unit with space-time attributes.

[0107] Extract the start state attribute node and the end state attribute node of each driving behavior associated edge in the driving behavior impact link, and map the start state attribute node and the end state attribute node to the time slice index and spatial grid coordinates of the spatiotemporal coding grid unit. For example, in a driving behavior associated edge, the start state attribute node is the engine speed node during rapid acceleration (located at a specific time slice index), and the end state attribute node is the brake pressure change node. Map these two nodes to the time slice index and spatial grid coordinates of the corresponding spatiotemporal coding grid unit. According to the time slice index of the spatiotemporal coding grid unit, slide and splice the continuous multiple driving behavior associated edges in the driving behavior impact link in the time window to generate a driving behavior propagation path across time slices. For example, the driving behavior associated edges in multiple continuous 5-minute time slices are spliced ​​in sequence to form a driving behavior propagation path across time, reflecting the propagation of driving behavior over a longer time span.

[0108] The transmission delay parameters of each communication interaction path are extracted from the communication delay risk propagation channel, and the driving behavior propagation path across time slices is delayed compensated and offset according to the transmission delay parameters to generate a delay-calibrated driving behavior propagation path. Assuming that the transmission delay of a certain communication interaction path is 2 seconds, this delay is taken into account when constructing the driving behavior propagation path, and the relevant driving behavior associated edges are adjusted in time to make the driving behavior propagation path more in line with the actual situation. Each node in the delay-calibrated driving behavior propagation path is matched with the spatial grid coordinates of the spatiotemporal coding grid unit to generate a driving behavior propagation path with spatial position constraints. In this way, each node in the driving behavior propagation path corresponds to the spatial position around the vehicle, clarifying the propagation of driving behavior in space.

[0109] The dynamic environmental risk factors of the environmental constraint layer are superimposed on the spatial grid coordinates of the spatiotemporal coding grid unit to generate the risk density weight of each spatiotemporal coding grid unit. For example, in a spatiotemporal coding grid unit, if the visibility attenuation coefficient of the area is high and the road adhesion risk index is large, then the risk density weight of this grid unit will be high. The risk transmission probability of the driving behavior propagation path with spatial position constraints is weighted according to the risk density weight to generate a risk transmission probability distribution path. If the driving behavior propagation path passes through a spatiotemporal coding grid unit with a high risk density weight, the risk transmission probability in this area will increase accordingly, thus forming a risk transmission probability distribution path.

[0110] Each node in the risk transmission probability distribution path is gradient-connected with the risk density weight of the adjacent spatiotemporal coding grid unit to generate a risk gradient transmission link. For example, the risk transmission probability of a certain node is affected by the risk density weight of the adjacent grid unit. The difference in the risk density weight of the adjacent grid unit forms the gradient of risk transmission, and in this way, a risk gradient transmission link is constructed. Time slice backtesting is performed on each node in the risk gradient transmission link to identify the set of nodes whose risk density weight exceeds the threshold in continuous time slices and generate a risk spatiotemporal focus area. If the risk density weight of the spatiotemporal coding grid unit where a node is located is always high and exceeds the set threshold in multiple continuous time slices, the node is included in the risk spatiotemporal focus area, indicating that this area is a high-risk area in both time and space.

[0111] According to the spatial grid coordinate distribution of the risk spatiotemporal focus area, multiple nodes located at the same spatial grid coordinate in the risk gradient transmission link are path aggregated to generate a spatial aggregate risk transmission link. For example, on the same spatial grid coordinate, risk-related nodes in different time slices are aggregated to form a spatial aggregate risk transmission link, which reflects the comprehensive situation of risk transmission at this spatial location. Each path node in the spatial aggregate risk transmission link is expanded in time dimension, and the path nodes of the same spatial grid coordinate in different time slices are connected as a time evolution risk transmission trajectory. In this way, the evolution of risk over time at the same spatial location can be seen.

[0112] The time-evolved risk transmission trajectory and the risk gradient transmission link are cross-fused to generate a multi-level risk transmission network with spatiotemporal evolution properties. The multi-level risk transmission network contains risk transmission relationships at different time and space levels, and fully reflects the spread of risks. Each node in the multi-level risk transmission network is dynamically mapped to the risk density weight of the spatiotemporal coding grid unit to generate a node risk weight label. A risk weight label is attached to each node according to the risk density weight of the node in different spatiotemporal coding grid units. The transmission paths in the multi-level risk transmission network are ranked according to the node risk weight labels to generate a priority-marked spatiotemporal correlation risk transmission map. The transmission paths where the nodes with higher risk weights are located have higher priorities.

[0113] Finally, the priority mark in the spatiotemporal correlation risk transmission map is spatially superimposed and verified with the dynamic environmental risk factors of the environmental constraint layer to eliminate the transmission paths with conflicting spatial grid coordinates. For example, if the priority mark of a transmission path conflicts with the actual risk situation in the environmental constraint layer (such as a certain area is actually low risk but the transmission path is displayed as high priority), the transmission path is adjusted. Based on the time slice division results of the spatiotemporal correlation risk transmission map after verification and the spatiotemporal coding grid unit, a spatiotemporal correlation risk analysis map with spatiotemporal stamps is generated, which accurately reflects the risk situation of the vehicle in the spatiotemporal dimension.

[0114] In a possible implementation, step S140 includes: Step S141, traverse the driving behavior impact links in the spatiotemporal correlation risk analysis graph, and calculate the risk transmission intensity of each link.

[0115] For example, in a driving behavior impact link, starting from the sudden acceleration behavior, through the engine speed mutation, brake pressure change and other nodes, the risk transmission relationship between each node has a certain strength. Sudden acceleration may cause the engine to suddenly bear a large load. When this load change is transmitted to the braking system, if the brake pressure cannot respond effectively and in time, it will increase the risk. The risk transmission strength of this link is quantified by analyzing the correlation between the nodes in the link, the value of the relevant feature vector, and environmental constraints. For example, when the engine speed mutation amplitude is large, the brake pressure response is delayed, and the environmental constraint is a slippery road surface (increasing the risk of braking distance), the risk transmission strength of this link will be high.

[0116] Step S142: determining a risk response hysteresis coefficient according to the path length and delay time of the communication delay risk propagation channel.

[0117] In the scenario where a vehicle communicates with a roadside unit or other vehicles, the path length of the communication delay risk propagation channel depends on factors such as the distance between the communication devices and the topology of the communication network. For example, when a vehicle communicates with a roadside unit that is farther away, the path is longer. At the same time, there may be a delay in communication. Assume that there is a 2-second delay in the vehicle obtaining traffic congestion information. Longer path lengths and larger delay times will lead to an increase in the risk response lag coefficient. This means that the driver will react later based on the information obtained through communication, increasing driving risks. For example, when a vehicle receives information that a sudden accident on the road ahead requires deceleration, if the driver delays braking for 2 seconds due to communication delays, the vehicle may have already approached the accident area during this period, greatly increasing the risk.

[0118] Step S143, generating an environmental risk superposition effect value based on the spatial grid density and risk factor weight of the environmental constraint layer.

[0119] The environmental constraint layer is composed of multiple factors, such as visibility attenuation coefficient, road adhesion risk index, etc. Taking the area around the vehicle divided into multiple spatial grid units as an example, there may be a high spatial grid density in some areas, that is, areas where environmental constraint factors are concentrated. For example, near an intersection, there are many traffic signs, a large vehicle density, and there may be road obstacles. The spatial grid density in this area is high. At the same time, each risk factor (such as the risk factor weight corresponding to the visibility attenuation coefficient, the risk factor weight corresponding to the road adhesion risk index, etc.) has a corresponding weight according to its impact on the risk. By comprehensively considering the spatial grid density and the weights of each risk factor, the environmental risk superposition effect value is calculated. In the example of the intersection, since multiple risk factors are concentrated and the weights are large, the environmental risk superposition effect value will be high.

[0120] Step S144, monitoring the real-time change rate of the characteristic vector of the state attribute node to generate a vehicle state degradation index.

[0121] The state attribute node contains relevant feature vectors such as engine speed sequence and brake pressure waveform. For example, for the engine speed sequence, its change rate is monitored in real time. If during normal driving, the engine speed suddenly fluctuates frequently and significantly, or the fluctuation amplitude gradually increases over time, this indicates that the operating state of the engine may be degrading. For the brake pressure waveform, if the response time of the brake pressure gradually becomes longer or the fluctuation of the brake pressure becomes unstable, it also indicates that there may be problems with the brake system. The vehicle state degradation index is generated by comprehensively analyzing the real-time change rate of the feature vectors of the above state attribute nodes. For example, abnormal changes in the engine speed and brake pressure waveforms lead to an increase in the vehicle state degradation index, indicating that the operating health of the vehicle's internal systems is declining, increasing driving risks.

[0122] Step S145, inputting the risk transmission intensity, the risk response hysteresis coefficient, the environmental risk superposition effect value and the vehicle state degradation index into a risk quantification model, and outputting the risk level quantification index.

[0123] The risk quantification model is a tool for accurately assessing risks based on historical data. The risk transmission intensity reflects the transmission risk of driving behavior in the vehicle's internal system and environment, the risk response lag coefficient reflects the impact of external communication on driving risk, the environmental risk superposition effect value represents the comprehensive risk impact of the external environment on the vehicle, and the vehicle state degradation index represents the health risk of the vehicle's own system. After these factors are input into the risk quantification model, the model outputs a risk level quantitative index based on its internal algorithms and parameters. For example, the risk level quantitative index calculated by the model is 70 (assuming a full score of 100), which means that the vehicle is currently at a high risk level.

[0124] Step S146, triggering risk event warning labels of different levels according to the comparison of the risk level quantitative index with a preset threshold.

[0125] If the preset threshold is 60, when the risk level quantitative index is 70, it exceeds the preset threshold and triggers the corresponding level of risk event warning label. For example, the risk event warning label of "medium-high risk-need close attention" may be triggered, prompting the driver of the vehicle or the relevant monitoring system that the vehicle is currently facing medium-high risks and needs to pay close attention to the vehicle status, driving behavior or environmental conditions or take corresponding measures.

[0126] In a possible implementation, the step of constructing the risk quantification model includes: Step S210, obtaining a sample set of risk events annotated in historical Internet of Vehicles data, and extracting a combined operation feature vector, an abnormal behavior pattern identifier, a dynamic environment risk factor, and a structured communication feature matrix corresponding to the sample.

[0127] In the historical driving data of electric vehicles, there are a large number of data that have been marked as risk events or normal events. For example, collision accidents caused by sudden acceleration and scratches caused by low visibility in bad weather are marked as risk events, while data during normal driving are marked as normal events. The combined operation feature vector, abnormal behavior pattern identifier, dynamic environmental risk factor and structured communication feature matrix corresponding to the sample are extracted from these samples. For risk event samples, the combined operation feature vector may contain features such as abnormal fluctuations in engine speed and unstable brake pressure; the abnormal behavior pattern identifier may show that the number of sudden acceleration events is high and the density of continuous lane change behavior is high; the dynamic environmental risk factor may include factors such as low visibility in rainy weather and slippery road surface; the structured communication feature matrix may reflect that there is a delay or interruption in communication before the accident.

[0128] Step S220, constructing an initial risk assessment neural network, wherein the initial risk assessment neural network includes a feature fusion layer, a temporal attention layer, and a spatial convolution layer.

[0129] Step S230: cross-modally fuse the combined operation feature vector and the dynamic environmental risk factor through the feature fusion layer to generate an environmental coupling feature.

[0130] The initial risk assessment neural network is a complex model built to accurately assess driving risks. The function of the feature fusion layer is to perform cross-modal fusion of the combined operation feature vector and the dynamic environmental risk factor to generate environmental coupling features. For example, the internal vehicle operation features such as engine speed and brake pressure are fused with external environmental features such as visibility and road adhesion. In this process, the features of different modes (internal vehicle state and external environment) are fused through a specific algorithm to generate environmental coupling features that can comprehensively reflect the state of the vehicle in the environment.

[0131] Step S240: Perform time-dependency modeling on the abnormal behavior pattern identifier through the time attention layer to generate a behavior timing weight.

[0132] In historical data, factors such as the number of sudden acceleration event triggers and the time interval of continuous lane change events in the abnormal behavior pattern identifier have different importance at different time points. For example, during traffic congestion, sudden acceleration events may have a higher risk weight than during normal driving. The temporal attention layer analyzes the changing rules of these abnormal behavior patterns in the time series and assigns corresponding behavior time series weights to the abnormal behavior patterns at each time point to reflect their risk importance at different times.

[0133] Step S250: extracting spatial features from the structured communication feature matrix through the spatial convolution layer to generate a communication risk distribution map.

[0134] The structured communication feature matrix contains the spatial relationship information between the vehicle and the surrounding devices. Through the convolution operation of the spatial convolution layer, the spatial characteristics of the communication can be extracted, such as the distribution of communication signal strength in different spatial locations, the area where communication interruption may occur, and other information, thereby generating a communication risk distribution map, which intuitively shows the distribution of communication risks in space.

[0135] Step S260, fully connect the environmental coupling characteristics, the behavior timing weights and the communication risk distribution map, and output a risk prediction value.

[0136] The fully connected layer comprehensively calculates the different features processed previously, and outputs a risk prediction value based on its internal connection weights and activation functions. This risk prediction value is a preliminary risk estimate based on the model's comprehensive evaluation of the input features.

[0137] Step S270, based on the back propagation of the error between the risk prediction value and the actual risk annotation value, optimize the parameters of the initial risk assessment neural network to obtain the risk quantification model.

[0138] During the initial risk assessment neural network training process, there is an error between the risk prediction value and the actual risk annotation value (the annotation of known risk events or normal events from historical data). Through the error back propagation algorithm, the error is back propagated from the output layer to the previous layers, and the parameters of each layer are adjusted (such as the fusion weight of the feature fusion layer, the time weight of the time attention layer, the convolution kernel parameters of the spatial convolution layer, etc.), and the model is continuously optimized, so that the error between the risk prediction value and the actual risk annotation value is gradually reduced, and finally an accurate risk quantification model is obtained.

[0139] In a possible implementation, step S150 includes: Step S151, determining the basic risk control strategy type according to the risk level quantitative index, including an emergency braking instruction, a speed limit control instruction, a path replanning instruction or a driver warning instruction.

[0140] In detail, when the risk level quantitative index reaches a certain value, the corresponding basic risk control strategy type needs to be adopted. For example, if the risk level quantitative index is high, it indicates that the vehicle faces a greater risk and an emergency braking command may need to be issued. When an electric vehicle is driving, if an obstacle is detected suddenly in front and the risk level quantitative index exceeds the threshold of emergency braking, it will be determined as an emergency braking command type to avoid collision risks. If the risk level quantitative index is at a medium level, it may be a speed limit control command. For example, when the road conditions are complex or the vehicle performance has certain abnormalities (such as excessive battery temperature), in order to ensure driving safety and vehicle system stability, the speed is limited to a reasonable range. When the risk level quantitative index shows that there are many risk factors in the surrounding environment (such as road construction and traffic congestion), a path replanning command may be triggered to guide the vehicle to choose a safer and smoother route. When the risk level quantitative index does not reach the emergency level, but there are some potential risk factors (such as initial signs of driver fatigue driving), a driver warning command will be issued to remind the driver to pay attention to safe driving.

[0141] Step S152, matching additional control parameters from the Internet of Vehicles strategy library in combination with the label type of the risk event warning label.

[0142] The risk event warning label contains more detailed risk information. For example, if the risk event warning label is "low visibility ahead - high risk", additional control parameters are matched from the Internet of Vehicles strategy library in combination with this label type. For emergency braking instructions, parameters for turning on the hazard warning lights and increasing the braking assist force at the same time when braking may be matched; for speed limit control instructions, parameters for further reducing the speed limit value according to visibility conditions may be matched, such as on a road with a normal speed limit of 60 kilometers per hour, due to low visibility ahead, the speed limit value is reduced to 30 kilometers per hour; for path replanning instructions, parameters for giving priority to roads with street lights or clear traffic signs as new paths will be matched; for driver warning instructions, additional control parameters such as reminding the driver to pay attention to low visibility ahead, drive carefully and turn on fog lights at the same time through voice and screen display will be matched.

[0143] Step S153, based on the real-time location data of the target vehicle and the high-precision map information, generate a strategy effectiveness range under geographical constraints.

[0144] Step S154, determining the executable instruction set and eliminating device conflicting instructions through hardware interface status detection of the vehicle-mounted control terminal.

[0145] The hardware interfaces of the vehicle control terminal include the braking system interface, the speed control system interface, the warning system interface, etc. Before issuing the risk control decision instruction, the status of these hardware interfaces must be detected. For example, check whether the braking system interface is working properly. If there is a fault in the braking system, the emergency braking instruction cannot be executed and needs to be excluded from the executable instruction set. For the speed control system interface, if it is performing other speed adjustment operations (such as the adaptive cruise system is running), it is necessary to check whether there is a conflict with the speed limit control instruction. If there is a conflict, it needs to be adjusted or eliminated. If there is a fault in the warning system interface, such as the voice warning function cannot work properly, then the voice reminder part of the driver's warning instruction needs to be adjusted or excluded, and only the feasible parts such as the screen display reminder are retained.

[0146] Step S155: logically combine the basic risk control strategy type, the additional control parameters, the strategy effective range and the executable instruction set to generate the risk control decision instruction set.

[0147] For example, the final risk control decision instruction set may be: within 200 meters ahead and within 10 minutes (strategy effectiveness range), if the vehicle speed is higher than 30 kilometers per hour (speed limit control instruction, additional control parameters), and the braking system interface is normal (executable instruction set), then the speed limit control instruction will be executed, and "The visibility ahead is low, please drive carefully" will be displayed on the screen (driver warning instruction, additional control parameters).

[0148] In a possible implementation, step S153 includes: Step S1531, obtaining the topological structure data and real-time traffic flow information of the road where the target vehicle is currently located.

[0149] Step S1532, extracting the road curvature radius, slope parameters and lane line type based on the high-precision map information.

[0150] Step S1533, calculating the vehicle density distribution and average driving speed in combination with the real-time traffic flow information.

[0151] Step S1534: generating a curve control radius based on the road curvature radius and the risk level quantitative index.

[0152] Step S1535: generating an avoidance area priority according to the vehicle density distribution and the risk event warning label.

[0153] Step S1536, integrating the curve control radius, the avoidance area priority and the lane line type, and defining the spatial boundary and time validity period of the strategy effective range.

[0154] Assume that the target vehicle is located on the main road of the city. The road topology data shows that this is a multi-lane road with multiple intersections. The real-time traffic flow information shows that the current traffic flow is large and the vehicle speed is relatively slow. The road curvature radius, slope parameters and lane line type are extracted based on the high-precision map information. If the road curvature radius is small, it means that there is a curve on the road and the slope is steep. The lane line type is a two-way four-lane with clear lane dividers and turn signs. Combined with real-time traffic flow information, the vehicle density distribution and average driving speed are calculated. In the case of large traffic flow, the vehicle density distribution is high and the average driving speed is low. Generate the curve control radius based on the road curvature radius and the risk level quantitative index. If the risk level quantitative index is high and the road curvature radius is small, in order to ensure the safety of the vehicle at the curve, it is necessary to generate a smaller curve control radius, such as further reducing the curve control radius corresponding to the curve safe passing speed originally designed according to the road, so as to reduce the vehicle speed at the curve and avoid the risk of skidding or loss of control. Generate the avoidance area priority based on the vehicle density distribution and risk event warning labels. If the risk event warning label is "small vehicle spacing-medium risk" and the vehicle density distribution is high, the avoidance area priority will be increased. For example, vehicles in the right lane will be given priority because the right lane may have more safety space or be more convenient for vehicles to perform avoidance operations. Integrate the curve control radius, avoidance area priority and lane line type to define the spatial boundary and time validity period of the strategy's effective range. For example, within the 200-meter road range ahead (spatial boundary) and within the next 10 minutes (time validity period), vehicles in these areas need to drive according to the curve control radius, and give priority to vehicle operations according to the avoidance area priority, while complying with the regulations of the lane line type.

[0155] In a possible implementation, the method further includes: Step S310, receiving the risk control decision instruction set of multiple target vehicles through the Internet of Vehicles cloud platform, and performing a global risk situation analysis.

[0156] Step S320, generating a regional risk heat map according to the geographical location distribution of the multiple target vehicles.

[0157] Step S330: adjusting the traffic signal control strategy and the roadside unit broadcast content based on the regional risk heat map.

[0158] Step S340: Send the adjusted traffic signal control strategy to the intelligent traffic management system and update the roadside unit broadcast content.

[0159] Step S350, generating a macro Internet of Vehicles risk trend report based on the aggregation results of the risk level quantitative indicators of the multiple target vehicles.

[0160] For example, the IoV cloud platform receives risk control decision instruction sets for multiple electric vehicles in different areas. Specifically, these vehicles are distributed in various corners of the city, some on main roads, some on branch roads, some on roads near commercial areas, and some on roads near residential areas. Generate a regional risk heat map based on the geographical location distribution of multiple target vehicles. Roads near commercial areas may be displayed as high-risk areas (red) on the regional risk heat map due to high vehicle density and complex traffic conditions; while roads near residential areas have relatively few vehicles and slow speeds, and may be displayed as low-risk areas (green). Adjust traffic signal control strategies and roadside unit broadcast content based on regional risk heat maps. For high-risk areas (such as roads near commercial areas), adjust traffic signal control strategies, such as extending red light time, reducing vehicle inflow, alleviating traffic congestion and reducing collision risks. At the same time, update the roadside unit broadcast content, such as broadcasting prompt information such as "The road ahead in the commercial area is congested, please drive carefully". The adjusted traffic signal control strategy is sent to the intelligent traffic management system so that the traffic signal operates according to the adjusted strategy, such as changing the duration setting of the signal light. And update the roadside unit broadcast content to inform drivers of relevant risk information and driving suggestions in a timely manner. Generate a macro-vehicle network risk trend report based on the aggregation results of the risk level quantitative indicators of multiple target vehicles. For example, if the risk level quantitative indicators of multiple vehicles show an upward trend over a period of time, it means that the risk of the entire vehicle network area is increasing. The macro-vehicle network risk trend report can analyze the increase in risk due to weather changes (such as reduced visibility and slippery roads due to rainy weather), peak traffic flow in a specific time period, road construction, etc., and put forward corresponding suggestions, such as increasing road patrols, strengthening traffic guidance and other measures to ensure traffic safety in the entire vehicle network area.

[0161] Figure 2 A schematic diagram of exemplary hardware and software components of a risk control big data mining system 100 based on the Internet of Vehicles that can implement the concept of the present application is shown in some embodiments of the present application. For example, the processor 120 can be used in the risk control big data mining system 100 based on the Internet of Vehicles and used to perform the functions in the present application.

[0162] The risk control big data mining system 100 based on the Internet of Vehicles can be a general server or a special-purpose server, both of which can be used to implement the risk control big data mining method based on the Internet of Vehicles of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0163] For example, the risk control big data mining system 100 based on the Internet of Vehicles may include a network port 110 connected to the network, one or more processors 120 for executing program instructions, a communication bus 130, and storage media 140 in different forms, such as disks, ROMs, or RAMs, or any combination thereof. Exemplarily, the risk control big data mining system 100 based on the Internet of Vehicles may also include program instructions stored in ROMs, RAMs, or other types of non-temporary storage media, or any combination thereof. The method of the present application may be implemented according to these program instructions. The risk control big data mining system 100 based on the Internet of Vehicles also includes an input / output (I / O) interface 150 between the computer and other input / output devices.

[0164] For ease of explanation, only one processor is described in the risk control big data mining system 100 based on the Internet of Vehicles. However, it should be noted that the risk control big data mining system 100 based on the Internet of Vehicles in the present application may also include multiple processors, so the steps performed by one processor described in the present application may also be performed jointly or individually by multiple processors. For example, if the processor of the risk control big data mining system 100 based on the Internet of Vehicles executes steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor execute steps A and B together.

[0165] 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 big data mining method based on the Internet of Vehicles is implemented.

[0166] 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 big data mining method based on Internet of Vehicles, characterized in that: The method comprises: Collecting real-time operation data streams generated by the target vehicle in the Internet of Vehicles environment, the real-time operation data streams including vehicle status time series data, driving behavior event sets, external environment perception data, and vehicle communication interaction records; extracting a combined operation feature vector from the vehicle state time series data, extracting an abnormal behavior pattern identifier from the driving behavior event set, parsing a dynamic environmental risk factor from the external environment perception data, and converting the vehicle communication interaction record into a structured communication feature matrix; Based on the combined operation feature vector, the abnormal behavior pattern identifier, the dynamic environment risk factor and the structured communication feature matrix, construct a spatiotemporal correlation risk analysis map; Performing a dynamic driving risk assessment on the target vehicle through the spatiotemporal correlation risk analysis graph to generate a risk level quantitative index and a risk event warning label; A risk control decision instruction set is generated according to the risk level quantitative index and the risk event warning label, and the risk control decision instruction set is fed back to the on-board control terminal of the target vehicle and the Internet of Vehicles cloud platform.

2. The risk control big data mining method based on the Internet of Vehicles according to claim 1 is characterized in that: The vehicle state time series data includes an engine speed sequence, a brake pressure waveform, a battery temperature curve, and a tire wear increment, and extracting a combined operation feature vector from the vehicle state time series data includes: Performing frequency domain transformation on the engine speed sequence to extract periodic fluctuation amplitude characteristics and speed mutation detection marks; Performing segmented integration processing on the brake pressure waveform to generate a brake intensity distribution histogram and a pressure abnormal fluctuation threshold; Performing trend decomposition on the battery temperature curve to separate long-term temperature rise gradient characteristics and short-term temperature oscillation parameters; Modeling the wear rate of the tire wear increment to calculate the tire life prediction index and the probability of abnormal tread wear; The periodic fluctuation amplitude characteristics, the speed mutation detection mark, the braking intensity distribution histogram, the pressure abnormal fluctuation threshold, the long-term temperature rise gradient characteristics, the short-term temperature oscillation parameters, the tire life prediction index and the tread abnormal wear probability are normalized and spliced ​​to form the combined operation feature vector.

3. The risk control big data mining method based on the Internet of Vehicles according to claim 1 is characterized in that: The extracting of abnormal behavior pattern identifiers from the driving behavior event set includes: The number of sudden acceleration event triggering, the time interval of continuous lane change events, the proportion of speeding duration, and the frequency of illegal light use at night are selected from the driving behavior event set; Generating a sudden acceleration risk coefficient based on the comparison of the sudden acceleration event triggering times with a preset acceleration threshold; determining a lane-changing behavior density score according to the statistical distribution of time intervals between the consecutive lane-changing events; Combining the speeding duration ratio with the road speed limit data to generate a speeding violation accumulation amount; Generate a lighting violation pattern identifier by matching the nighttime illegal lighting usage frequency with the lighting specification database; The sudden acceleration risk coefficient, the lane change behavior density score, the speeding violation accumulation amount and the light violation pattern identifier are weightedly fused to generate the abnormal behavior pattern identifier.

4. The risk control big data mining method based on the Internet of Vehicles according to claim 1 is characterized in that: The step of analyzing the dynamic environmental risk factors from the external environmental perception data includes: Identify real-time meteorological condition parameters, road obstacle distribution heat map, traffic sign semantic information and surrounding vehicle movement trajectory set from the external environment perception data; Calculating the visibility attenuation coefficient and the road adhesion risk index according to the real-time meteorological condition parameters; Generating an obstacle collision probability distribution map based on the density gradient of the road obstacle distribution heat map; Performing compliance verification on the traffic sign semantic information and generating a sign violation risk weight; Performing relative speed cluster analysis on the surrounding vehicle motion trajectory set to generate a vehicle spacing safety assessment matrix; The visibility attenuation coefficient, the road adhesion risk index, the obstacle collision probability distribution map, the sign violation risk weight and the vehicle spacing safety assessment matrix are spatially superimposed to generate the dynamic environment risk factor.

5. The risk control big data mining method based on Internet of Vehicles according to claim 1 is characterized in that: The constructing of a spatiotemporal correlation risk analysis graph based on the combined operation feature vector, the abnormal behavior pattern identifier, the dynamic environment risk factor and the structured communication feature matrix includes: Mapping the combined operation feature vector to the state attribute node of the target vehicle, mapping the abnormal behavior pattern identifier to the driving behavior association edge, mapping the dynamic environment risk factor to the environment constraint layer, and mapping the structured communication feature matrix to the communication interaction path; Performing time slicing division on the state attribute nodes based on a preset time window to generate a time dimension topological relationship; Performing spatial grid division on the environmental constraint layer according to the vehicle geographic location data to generate a spatial dimension association matrix; Connecting the state attribute nodes in the same time slice through the driving behavior association edge to form a driving behavior influence link; Cross-layer coupling of the communication interaction path and the driving behavior impact link to generate a communication delay risk propagation channel; The time dimension topological relationship, the space dimension association matrix, the driving behavior impact link and the communication delay risk propagation channel are integrated to form the time-space association risk analysis map.

6. The risk control big data mining method based on the Internet of Vehicles according to claim 5 is characterized in that: The dynamic evaluation of the driving risk of the target vehicle is performed through the spatiotemporal correlation risk analysis graph to generate a risk level quantitative index and a risk event warning label, including: Traversing the driving behavior impact links in the spatiotemporal correlation risk analysis graph, and calculating the risk transmission intensity of each link; Determining a risk response hysteresis coefficient according to the path length and delay time of the communication delay risk propagation channel; Generate an environmental risk superposition effect value based on the spatial grid density and risk factor weights of the environmental constraint layer; Performing real-time change rate monitoring on the characteristic vector of the state attribute node to generate a vehicle state degradation index; Input the risk transmission intensity, the risk response hysteresis coefficient, the environmental risk superposition effect value and the vehicle state degradation index into a risk quantification model, and output the risk level quantification index; According to the comparison between the risk level quantitative index and the preset threshold, the risk event warning labels of different levels are triggered.

7. The risk control big data mining method based on the Internet of Vehicles according to claim 6 is characterized in that: The steps of constructing the risk quantification model include: Obtain risk event sample sets annotated in historical Internet of Vehicles data, and extract the combined operation feature vectors, abnormal behavior pattern identifiers, dynamic environment risk factors, and structured communication feature matrices corresponding to the samples; Constructing an initial risk assessment neural network, wherein the initial risk assessment neural network includes a feature fusion layer, a temporal attention layer, and a spatial convolution layer; Cross-modally fusing the combined operation feature vector and the dynamic environmental risk factor through the feature fusion layer to generate an environmental coupling feature; Modeling the time dependency of the abnormal behavior pattern identifier through the time attention layer to generate a behavior timing weight; Performing spatial feature extraction on the structured communication feature matrix through the spatial convolution layer to generate a communication risk distribution map; Performing full connection processing on the environmental coupling feature, the behavior timing weight and the communication risk distribution map, and outputting a risk prediction value; Based on the back propagation of the error between the risk prediction value and the actual risk annotation value, the parameters of the initial risk assessment neural network are optimized to obtain the risk quantification model.

8. The risk control big data mining method based on the Internet of Vehicles according to claim 1 is characterized in that: The generating of the risk control decision instruction set according to the risk level quantitative index and the risk event warning label includes: Determine the type of basic risk control strategy based on the risk level quantitative indicators, including emergency braking instructions, speed limit control instructions, path re-planning instructions or driver warning instructions; In combination with the label type of the risk event warning label, additional control parameters are matched from the Internet of Vehicles strategy library; Based on the real-time location data of the target vehicle and high-precision map information, generate a strategy effective range under geographical constraints; Determine the executable instruction set and eliminate device conflicting instructions through hardware interface status detection of the vehicle control terminal; Logically combining the basic risk control strategy type, the additional control parameters, the strategy effective range, and the executable instruction set to generate the risk control decision instruction set; The strategy effectiveness range under the geographical constraint condition is generated based on the real-time location data of the target vehicle and the high-precision map information, including: Obtaining topological structure data and real-time traffic flow information of the road where the target vehicle is currently located; Extracting road curvature radius, slope parameters and lane line type according to the high-precision map information; Calculate vehicle density distribution and average driving speed in combination with the real-time traffic flow information; Generating a curve control radius based on the road curvature radius and the risk level quantitative index; generating an avoidance area priority according to the vehicle density distribution and the risk event warning label; The curve control radius, the avoidance area priority and the lane line type are integrated to define the spatial boundary and time validity period of the strategy effective range.

9. The risk control big data mining method based on the Internet of Vehicles according to claim 1 is characterized in that: The method further comprises: Receive risk control decision instruction sets of multiple target vehicles through the Internet of Vehicles cloud platform and conduct global risk situation analysis; Generating a regional risk heat map according to the geographical location distribution of the multiple target vehicles; Adjusting the traffic signal control strategy and the roadside unit broadcast content based on the regional risk heat map; Sending the adjusted traffic signal control strategy to the intelligent traffic management system and updating the roadside unit broadcast content; Based on the aggregation results of the quantitative indicators of the risk levels of the multiple target vehicles, a macro Internet of Vehicles risk trend report is generated.

10. A risk control big data mining system based on Internet of Vehicles, characterized in that: The risk control big data mining 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 big data mining method based on the Internet of Vehicles as described in any one of claims 1 to 9.

Images