Authority safety control method and device for artificial intelligence automobile

Abstract

The invention discloses an artificial intelligence automobile authority safety control method and device, and relates to the technical field of automobile electronics. According to the technical scheme, physiological monitoring data, driving behavior data and environment sensing data in a current period are acquired; calculating a physiological anomaly index, an operation deviation degree and an environment danger coefficient of the current period; inputting the physiological anomaly index, the operation deviation degree and the environmental risk coefficient into a risk prediction model fused with an attention mechanism to obtain a dynamic risk value of a next period; correcting the dynamic risk value according to the risk aging decay factor, and matching a target risk level according to the corrected dynamic risk value; determining a dynamic permission adjustment strategy corresponding to the target risk level by using a risk-permission linkage rule; and performing real-time verification on the dynamic permission adjustment strategy based on the driving scene where the vehicle is located, generating a controllable function list, and executing permission authorization operation in the next period. The main purpose is to realize accurate permission adaptation of vehicle functions.

Description

Technical Field

[0001] This application relates to the field of automotive electronics technology, and particularly to a method and device for permission security control of an intelligent vehicle with artificial intelligence. Background Art

[0002] With the rapid development of artificial intelligence and autonomous driving technologies, intelligent vehicles are gradually evolving from single driving assistance functions to a "co-driving mode" of human-machine collaboration. In this scenario, how to dynamically balance the driver's control right and the vehicle's autonomous decision-making power has become the core challenge in ensuring driving safety. Especially in high-risk scenarios such as sudden illness, fatigue driving, and complex road conditions, it may lead to driver misoperations or system response lags, and further trigger safety accidents.

[0003] Currently, most of the permission security controls for vehicles in the prior art adopt fixed permission rules for single-dimensional data sources. For example, only the driver's state is detected through the steering wheel grip force, or only lane departure is identified through a camera. However, although this method can achieve partial function takeover, it lacks the collaborative analysis of physiological indicators, operation behaviors, and environmental risks, and it is difficult to comprehensively quantify dynamic risks, resulting in the vehicle's permission management mechanism being unable to perceive multi-dimensional risks in real time and flexibly adjust vehicle function permissions, severely restricting the safety redundancy and scenario adaptability of vehicle permission control, and affecting the collaborative safety and control flexibility of vehicle driving. Summary of the Invention

[0004] In view of the above problems, this application provides a method and device for permission security control of an intelligent vehicle with artificial intelligence. The main purpose is to predict the comprehensive risk value in the future period through multi-modal data of physiology, behavior, and environment, and to achieve precise permission adaptation of vehicle functions based on the dynamic risk level, so as to improve the collaborative safety and control flexibility of vehicle driving.

[0005] To solve the above technical problems, this application proposes the following solutions: In a first aspect, this application provides a method for permission security control of an intelligent vehicle with artificial intelligence, and the method includes: Obtain the physiological monitoring data, driving behavior data of the user, and environmental perception data of the vehicle in the current period; Perform spatio-temporal alignment and noise filtering processing on the physiological monitoring data, the driving behavior data, and the environmental perception data, and calculate the physiological abnormality index, operation deviation degree, and environmental danger coefficient of the current period respectively; Input the physiological abnormality index, operation deviation degree, and environmental hazard coefficient of the current cycle into the risk prediction model with a fusion attention mechanism to obtain the dynamic risk value of the next cycle. The attention mechanism is used to dynamically allocate the global attention weights of the physiological abnormality index, the operation deviation degree, and the environmental hazard coefficient according to the risk evolution relationship of historical cycles; Correct the dynamic risk value of the next cycle according to the risk time decay factor, and match the corrected dynamic risk value with the target risk level in the preset risk levels. The risk time decay factor is determined by combining the dynamic risk values of historical cycles and the real-time state parameters of the vehicle; Use the risk-privilege linkage rule to determine the dynamic privilege adjustment strategy corresponding to the target risk level. The dynamic privilege adjustment strategy includes the hierarchical authorization threshold of function privileges and the real-time response rule; Based on the driving scenario where the vehicle is located, perform real-time verification on the dynamic privilege adjustment strategy, generate a list of finally authorized controllable functions, and perform privilege authorization operations in the next cycle for the user to perform safe control.

[0006] In a second aspect, the present application provides a privilege safety control device for an artificial intelligence vehicle, which is applied to the privilege safety control method of the artificial intelligence vehicle in the first aspect above. The device includes: A first acquisition unit, configured to acquire the physiological monitoring data, driving behavior data of the user, and environmental perception data of the vehicle in the current cycle; A first calculation unit, configured to perform spatio-temporal alignment and noise filtering processing on the physiological monitoring data, the driving behavior data, and the environmental perception data, and respectively calculate the physiological abnormality index, operation deviation degree, and environmental hazard coefficient of the current cycle; A prediction unit, configured to input the physiological abnormality index, operation deviation degree, and environmental hazard coefficient of the current cycle into the risk prediction model with a fusion attention mechanism to obtain the dynamic risk value of the next cycle. The attention mechanism is used to dynamically allocate the global attention weights of the physiological abnormality index, the operation deviation degree, and the environmental hazard coefficient according to the risk evolution relationship of historical cycles; A processing unit, configured to correct the dynamic risk value of the next cycle according to the risk time decay factor, and match the corrected dynamic risk value with the target risk level in the preset risk levels. The risk time decay factor is determined by combining the dynamic risk values of historical cycles and the real-time state parameters of the vehicle; A determination unit, configured to use the risk-privilege linkage rule to determine the dynamic privilege adjustment strategy corresponding to the target risk level. The dynamic privilege adjustment strategy includes the hierarchical authorization threshold of function privileges and the real-time response rule; An authorization unit, configured to perform real-time verification on the dynamic permission adjustment policy based on the driving scenario in which the vehicle is located, generate a list of finally authorized controllable functions, and perform a permission authorization operation in the next cycle for the user to perform safe control.

[0007] To achieve the above object, according to a third aspect of the present application, there is provided a storage medium, the storage medium including a stored program, wherein when the program runs, it controls the device where the storage medium is located to execute the permission security control method of the artificial intelligence vehicle in the first aspect above.

[0008] To achieve the above object, according to a fourth aspect of the present application, there is provided a processor, the processor being used to run a program, wherein when the program runs, it executes the permission security control method of the artificial intelligence vehicle in the first aspect above.

[0009] With the above technical solution, a method and device for permission security control of an artificial intelligence vehicle provided by the present application first obtain the physiological monitoring data, driving behavior data of the user, and environmental perception data of the vehicle in the current cycle, and then perform spatio-temporal alignment and noise filtering processing on the physiological monitoring data, driving behavior data, and environmental perception data, and calculate the physiological abnormality index, operation deviation degree, and environmental danger coefficient of the current cycle respectively. Then, the physiological abnormality index, operation deviation degree, and environmental danger coefficient of the current cycle are input into the risk prediction model with a fusion attention mechanism to obtain the dynamic risk value of the next cycle. The attention mechanism is used to dynamically allocate the global attention weights of the physiological abnormality index, operation deviation degree, and environmental danger coefficient according to the risk evolution relationship of the historical cycle. Then, the dynamic risk value of the next cycle is corrected according to the risk time decay factor, and the corrected dynamic risk value is matched with the target risk level in the preset risk level. The risk time decay factor is determined by combining the dynamic risk value of the historical cycle and the real-time state parameters of the vehicle. Subsequently, the dynamic permission adjustment strategy corresponding to the target risk level is determined by using the risk-permission linkage rule. The dynamic permission adjustment strategy includes the hierarchical authorization threshold of the function permission and the real-time response rule. Finally, the dynamic permission adjustment strategy is verified in real time based on the driving scenario where the vehicle is located, a list of manipulable functions with final authorization is generated, and the permission authorization operation is performed in the next cycle to enable the user to perform security control.The technical solution provided by this application synchronously collects physiological monitoring data, driving behavior data, and environmental perception data in the current cycle, calculates the physiological abnormality index, operation deviation degree, and environmental risk coefficient respectively, breaks through the limitations of single-dimensional risk assessment, realizes the comprehensive-factor risk coupling analysis of "human-vehicle-environment", avoids misjudgment caused by isolated judgment in traditional solutions, improves the comprehensiveness and accuracy of subsequent risk prediction, inputs the physiological abnormality index, operation deviation degree, and environmental risk coefficient in the current cycle into a risk prediction model that dynamically assigns global attention weights to the physiological abnormality index, operation deviation degree, and environmental risk coefficient according to the risk evolution relationship in historical cycles, outputs the dynamic risk value in the next cycle, can capture the risk evolution trend, realizes pre-risk perception, enables permission adjustment to be ahead of the risk outbreak, corrects the dynamic risk value based on the risk time-effect decay factor, and matches the preset risk level after correction, considering the characteristic that the impact of risk events decays over time, making the matching of risk levels more accurate, generates a dynamic permission adjustment strategy through risk-permission linkage rules, and performs real-time verification on the dynamic permission adjustment strategy based on the driving scenario where the vehicle is located, generates a list of finally authorized controllable functions, can ensure the compatibility between the list of controllable functions and the driving scenario, thus breaking the rigid restrictions of static permission rules, realizing elastic expansion and contraction of permissions, while ensuring core safety functions, avoiding excessive restriction of reasonable operations by drivers, balancing safety and driving freedom, and continuously updating the "human-vehicle-environment" data, dynamic risk value, and list of controllable functions in units of cycles, forming a "perception-prediction-decision-execution" closed loop, can adapt to the continuous changes in long-duration driving scenarios, thereby realizing real-time perception of multi-dimensional risks and flexibly adjusting vehicle function permissions, improving the safety redundancy and scenario adaptability of vehicle permission control, and further ensuring the collaborative safety and control flexibility of vehicle driving.

[0010] The above description is only an overview of the technical solution of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the description. And in order to make the above and other purposes, features, and advantages of this application more obvious and understandable, the following specifically illustrates the specific implementation manners of this application. Brief Description of the Drawings

[0011] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of this application. And throughout the drawings, the same reference numerals are used to represent the same components. In the drawings: Figure 1 Shows a flowchart of a method for controlling the permission safety of an artificial intelligence vehicle provided by an embodiment of this application; Figure 2Shows a flowchart of another method for controlling the permission security of an AI vehicle provided by an embodiment of the present application; Figure 3 Shows a block diagram of the composition of a permission security control device for an AI vehicle provided by an embodiment of the present application; Figure 4 Shows a block diagram of the composition of another permission security control device for an AI vehicle provided by an embodiment of the present application. Detailed implementation manners

[0012] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present application can be more thoroughly understood and the scope of the present application can be fully conveyed to those skilled in the art.

[0013] Currently, most of the permission security controls for vehicles in the prior art adopt fixed permission rules for a single-dimensional data source. For example, the driver's state is detected only by the steering wheel grip force, or lane departure is recognized only by a camera. However, although this method can achieve partial function takeover, it lacks the collaborative analysis of physiological indicators, operation behaviors, and environmental risks, and it is difficult to comprehensively quantify dynamic risks, resulting in the vehicle's permission management mechanism being unable to perceive multi-dimensional risks in real time and flexibly adjust vehicle function permissions, severely restricting the safety redundancy and scenario adaptability of vehicle permission control, and affecting the collaborative safety and control flexibility of vehicle driving.

[0014] Therefore, an embodiment of the present application provides a method for controlling the permission security of an AI vehicle. Through this method, the comprehensive risk value in the future period can be predicted through multi-modal data of physiology, behavior, and environment, and the precise permission adaptation of vehicle functions can be achieved based on the dynamic risk level, thereby improving the collaborative safety and control flexibility of vehicle driving. The specific implementation steps are as Figure 1 shown and include: 101. Obtain the physiological monitoring data, driving behavior data of the user, and environmental perception data of the vehicle in the current period.

[0015] In this step, the current cycle can be determined based on the collection requirements of physiological monitoring data, driving behavior data, and vehicle environmental perception data and the real-time requirements of risk prediction, for example, 30 seconds, 1 minute, 5 minutes, etc. Physiological monitoring data includes but is not limited to the user's breathing rate, micro-tremor amplitude, sitting stability, and the proportion of hand contact time with the steering wheel. Driving behavior data includes but is not limited to the accelerator and brake pedal pressure gradient, steering angular velocity, and vehicle acceleration data. Environmental perception data includes but is not limited to obstacle information, lane line status information, traffic flow information, and traffic event information.

[0016] Specifically, the detailed execution process of obtaining the user's physiological monitoring data, driving behavior data and the vehicle's environmental perception data in the current cycle is as follows: collecting the user's breathing rate, micro-tremor amplitude, sitting posture stability and the proportion of time that the hand contacts the steering wheel in the current cycle to obtain the user's physiological monitoring data in the current cycle; collecting the accelerator pedal pressure gradient, brake pedal pressure gradient, steering angular velocity and vehicle acceleration in the current cycle to obtain the user's driving behavior data in the current cycle; obtaining obstacle information, lane line status information, traffic flow information and traffic event information in the radar point cloud data of the vehicle's surrounding environment in the current cycle to obtain the vehicle's environmental perception data in the current cycle.

[0017] It should be noted that in the process of collecting physiological monitoring data, for the user's breathing frequency and micro-tremor amplitude, a non-contact millimeter-wave radar can be installed on the top of the car to collect the chest rise and fall frequency in real time to calculate the user's breathing frequency, and to calculate the micro-tremor amplitude through high-frequency signal analysis of small vibrations of the hands or head. For sitting posture stability, the smart seat pressure sensor can be embedded in the seat to collect the user's sitting posture stability, sitting posture pressure balance, sitting posture center of gravity offset distance, etc. in real time to calculate the sitting posture stability. As for the proportion of the time that the hand is in contact with the steering wheel, the capacitance change of the steering wheel capacitance sensor can be used to determine whether the hand has left the steering wheel, and the proportion of the hand contact time can be determined based on the proportion of the time that the hand is in contact with the steering wheel per unit time.

[0018] In the process of collecting driving behavior data, for the accelerator and brake pedal pressure gradient, high-precision pressure sensors can be built into the accelerator and brake pedals to determine the accelerator / brake pedal pressure gradient by calculating the pressure change rate per unit time. For the steering angular velocity, the steering angular velocity of the steering wheel (unit: ° / s) can be collected through the steering angle sensor. For vehicle acceleration, the longitudinal / lateral acceleration can be collected in real time by installing a vehicle acceleration sensor on the chassis.

[0019] During the process of collecting environmental perception data, for obstacle information, radar point cloud data can be collected through in-vehicle radar and cameras to generate obstacle information, such as the number of obstacles, the distance to the obstacles, the relative speed to the obstacles, and the type of obstacles. For lane line status information, the integrity of the lane lines can be calculated based on camera image recognition (such as intact, slightly damaged, severely damaged, etc.) and curve coordinate system transformation to obtain lane line status information, such as intact, slightly damaged, severely damaged, etc. For traffic flow information, the density of the traffic flow ahead, the occupancy rate of the road by vehicles, and the difference between the average speed of the vehicle and the speed limit can be obtained through in-vehicle cameras and vehicle-to-everything (V2X) communication to obtain traffic flow information. For traffic event information, traffic event information can be received in real time through V2X communication, such as accidents, construction, sudden weather changes, etc.

[0020] 102. Perform spatio-temporal alignment and noise filtering on physiological monitoring data, driving behavior data, and environmental perception data, and calculate the physiological abnormality index, operation deviation degree, and environmental danger coefficient for the current period respectively.

[0021] In this step, after obtaining the above-mentioned physiological monitoring data, driving behavior data, and environmental perception data, all sensor data are aligned to a unified time reference (error < 10ms) through GPS timestamps to complete time synchronization. Then, the radar point cloud data and camera images are mapped to the vehicle coordinate system to achieve coordinate system transformation, and different modality data (such as respiratory rate and steering angular velocity) are aligned on the time axis through the dynamic time warping (DTW) algorithm to achieve event association, and then spatial alignment is completed to obtain spatio-temporally aligned data. Noise filtering is performed on the spatio-temporally aligned data, including Kalman filtering and outlier detection, etc. Kalman filtering is used to smooth continuous data such as acceleration and steering angle, and outlier detection is based on the 3σ criterion to remove outliers in the sensor data (such as sudden acceleration mutations) to ensure data accuracy.

[0022] The physiological abnormality index is calculated based on the user's respiratory rate, micro-tremor amplitude, sitting posture stability, and the proportion of the contact duration between the hand and the steering wheel. The operation deviation degree is calculated based on the pressure gradients of the accelerator and brake pedals, the steering angular velocity, and the vehicle acceleration. The environmental danger coefficient is calculated based on the obstacle information, lane line status information, traffic flow information, and traffic event information.

[0023] Specifically, the detailed execution process of performing spatio-temporal alignment and noise filtering on physiological monitoring data, driving behavior data, and environmental perception data, and calculating the physiological abnormality index, operation deviation degree, and environmental danger coefficient for the current cycle respectively is as follows: Perform spatio-temporal alignment and noise filtering on physiological monitoring data, driving behavior data, and environmental perception data to obtain the processed physiological monitoring data, driving behavior data, and environmental perception data; Calculate the abnormal score of respiratory rate, abnormal score of micro-tremor amplitude, abnormal score of sitting posture stability, and abnormal score of contact duration based on the processed physiological monitoring data respectively, and determine the physiological abnormality index according to the abnormal score of respiratory rate, abnormal score of micro-tremor amplitude, abnormal score of sitting posture stability, and abnormal score of contact duration; Calculate the deviation degree of throttle pressure gradient, deviation degree of brake pressure gradient, deviation degree of steering angular velocity, and deviation degree of vehicle acceleration based on the processed driving behavior data respectively, and determine the operation deviation degree according to the deviation degree of throttle pressure gradient, deviation degree of brake pressure gradient, deviation degree of steering angular velocity, and deviation degree of vehicle acceleration; Calculate the obstacle danger level, lane line integrity, traffic flow danger level, and event urgency based on the processed environmental perception data respectively, and determine the environmental danger coefficient according to the obstacle danger level, lane line integrity, traffic flow danger level, and event urgency.

[0024] It should be noted that for the physiological abnormality index of the current cycle, the specific calculation process and expression are as follows: Convert the physiological indicators such as the respiratory rate, micro-tremor amplitude, sitting posture stability, and the proportion of the contact duration between the hand and the steering wheel in the current cycle into standardized scores (0 - 1) to characterize the degree of deviation from the normal range: Abnormal score of respiratory rate R: ; Where, if R > 1, then R = 1 (the maximum abnormal value).

[0025] Abnormal score of micro-tremor amplitude T: ; Where, if T < 0, then T = 0 (normal range); if T > 1, then T = 1.

[0026] Abnormal score of sitting posture stability S: Calculate the standard deviation of the pressure in the front / back and left / right areas of the seat: ; Where, is the pressure value of each pressure sensor, is the mean value, and N is the total number of pressure sensors embedded in the seat, which is used to characterize the uniformity of the seat pressure distribution.

[0027] ; Among them, if S > 1, then S = 1, which indicates that the comprehensive situation of the center of gravity offset and pressure balance of vehicle occupants has exceeded the safety threshold and is in a dangerous and attention-required state. When performing applications such as permission control and safety warning based on S later, as long as S = 1, the corresponding high-risk handling process can be triggered.

[0028] Abnormal fraction H of hand contact duration: ; If the contact duration ratio < 80%, the value of H increases.

[0029] ; Among them, if H > 1, then H = 1 (fully detached), which represents that the hand contact duration seriously does not meet the safety requirements and is in a dangerous situation such as "fully detached". In the safety control of vehicle driving, once H > 1, an emergency safety response can be triggered to ensure the safety and standardization of personnel operations during driving.

[0030] Corresponding weights are set for the abnormal fraction of respiratory rate, abnormal fraction of micro-tremor amplitude, abnormal fraction of sitting posture stability, and abnormal fraction of hand contact duration in advance according to the influence degree of the above physiological indicators on the safe driving of users. For example, respiratory rate (0.3), micro-tremor amplitude (0.25), sitting posture stability (0.25), hand contact duration (0.3). Then the current physiological abnormality index X: .

[0031] For the operation deviation degree in the current cycle, the specific calculation process and expression are: Throttle pedal pressure gradient deviation degree and brake pedal pressure gradient deviation degree : Reference value: Set the safety threshold according to the vehicle model (for example, the throttle pedal pressure gradient corresponding to the emergency acceleration threshold of 5 m / s², and the pedal pressure gradient corresponding to the emergency braking deceleration threshold of 20 m / s²).

[0032]

[0033] Among them, if the deviation degree exceeds 100%, then take 100% (maximum abnormality).

[0034] Steering angular velocity deviation degree : Reference value: Set the safety threshold according to the road type (such as highway / urban) (for example, the steering angular velocity threshold for urban roads is 300° / second).

[0035] ; Among them, if the measured value exceeds the safety threshold, take 1.0 (maximum anomaly).

[0036] Vehicle acceleration deviation degree : Reference value: Set the safe acceleration range based on the road speed limit and driving scenario (such as the safe range of longitudinal acceleration ±3 m / s²).

[0037] ; Among them, if the deviation degree exceeds 3 times the standard deviation, take the value corresponding to 3 times the standard deviation (such as 3.0).

[0038] Set corresponding weights for the deviation degrees of the throttle pedal pressure gradient, brake pedal pressure gradient, steering angular velocity deviation degree, and vehicle acceleration deviation degree in advance according to the influence degree of the above deviation indicators on the user's safe driving. For example, throttle pedal (0.3), brake pedal (0.4), steering angular velocity (0.2), vehicle acceleration (0.1). Then the current operation deviation degree Y: .

[0039] For the environmental danger coefficient of the current cycle, the specific calculation process and expression are: Obstacle danger degree : ; Among them, N is the number of detected obstacles; is the distance of obstacle i; is the relative speed with the vehicle ; is the obstacle type weight (such as pedestrian = 1.2, vehicle = 0.8, etc.).

[0040] Lane line state : .

[0041] Traffic flow danger degree : ; Among them, is the road speed limit; is the average speed; Occpancy is the proportion of the road occupied by vehicles (usually expressed as a percentage), including the occupancy of lanes or sections; is the traffic flow density; if <20% , then Weighted doubling (congestion state).

[0042] V2X event impact weighting: Event urgency : Set weights according to V2X event types (e.g., accident = 1.5, congestion = 0.8, construction = 0.6) ; Among them, is the event type, is the straight-line distance from the current vehicle position to the V2X event occurrence location, is the event impact distance threshold (e.g., 100 meters).

[0043] Current environmental danger coefficient Z: .

[0044] Based on the above detailed implementation methods, through the physiological abnormality index, the physiological state of the driver (such as breathing rate, micro-tremor, sitting posture stability, etc.) can be converted into a numerical risk indicator, reflecting the degree of fatigue, distraction or health abnormality. Through the operation deviation degree, driving behavior data (such as sudden acceleration, sudden braking, sudden change in steering angle, etc.) can be analyzed to quantify the deviation degree of the driver's operation from the "normal driving template" and identify dangerous driving behaviors. Through the environmental danger coefficient, environmental perception data (such as obstacle distance, lane line damage, traffic congestion, etc.) can be integrated to evaluate the potential danger degree of the surrounding environment. Realize the risk coupling analysis of all elements of "human-vehicle-environment", avoid misjudgment caused by isolated judgment due to single-dimensional data sources in traditional solutions, and effectively improve the comprehensiveness and accuracy of subsequent risk prediction.

[0045] 103. Input the physiological abnormality index, operation deviation degree, and environmental danger coefficient of the current cycle into the risk prediction model with a fusion attention mechanism to obtain the dynamic risk value of the next cycle.

[0046] Among them, the attention mechanism is used to dynamically allocate global attention weights to the physiological abnormality index, operation deviation degree, and environmental danger coefficient according to the risk evolution relationship of historical cycles.

[0047] In this step, a spatio-temporal convolutional network is used to process the spatio-temporal dependencies of the physiological abnormality index, the operation deviation degree, and the environmental danger coefficient, and the global attention weights of the physiological abnormality index, the operation deviation degree, and the environmental danger coefficient are dynamically allocated according to the attention mechanism. For example, when the event type is an emergency event, the weight of the environmental danger coefficient is increased, etc. The risk prediction model is specifically an LSTM-Transformer hybrid network model. The LSTM layer is used to capture the risk evolution relationship between the physiological abnormality index, the operation deviation degree, and the environmental danger coefficient in the historical period and the risk level. The Transformer layer is used to calculate the global attention weights of the physiological abnormality index, the operation deviation degree, and the environmental danger coefficient according to the risk evolution relationship. This risk prediction model is used to output the dynamic risk value of the next period after inputting the physiological abnormality index, the operation deviation degree, and the environmental danger coefficient of the current period, that is, its output layer is the dynamic risk value ∈[0,1], where: <0.3: low risk; 0.3 ≤ <0.7: medium risk; ≥0.7: high risk.

[0048] Training data is constructed based on the historical data set. This historical data set can specifically be the historical driving data of the past 1 or 3 years. Using the same processing method as in steps 101-102, that is, the historical driving data is divided into historical periods according to the time span of the current period, and the corresponding physiological abnormality index, environmental danger coefficient, and operation deviation degree of each historical period are calculated. The physiological abnormality index, environmental danger coefficient, and operation deviation degree are used as training data, and risk value annotation is performed in combination with the real risk event record to form training samples. The real risk event record at least includes the event type (such as collision, sudden braking, lane departure), the occurrence timestamp, and the severity (such as high, medium, low), etc. A dynamic time window can be set in advance for each event type, and according to the time window, the occurrence timestamp is combined with the risk event for matching and association. For example, the window length of an emergency event (collision) is 1-5 seconds. Comparing the occurrence timestamp, if there is a risk event within this time window, the risk value is marked according to the severity corresponding to this risk event. If there are multiple risk events within this time window, the highest severity among the multiple risk events is taken for risk value marking.

[0049] Construct an LSTM-Transformer hybrid network as a risk prediction model integrating an attention mechanism. The LSTM layer is used to capture the risk evolution relationship between the physiological anomaly index, operation deviation degree, environmental hazard coefficient, and risk level in historical periods. The Transformer layer is used to calculate the global attention weights of the physiological anomaly index, operation deviation degree, and environmental hazard coefficient based on the risk evolution relationship. For the training samples, they can be divided into a training set (70%), a validation set (15%), and a test set (15%). Use the training set to train the above-constructed LSTM-Transformer hybrid network model, use the validation set to verify the accuracy of the trained LSTM-Transformer hybrid network model, and then use the test set to test the performance of the verified LSTM-Transformer hybrid network model to make it meet the preset accuracy standard of the LSTM-Transformer hybrid network model, thereby obtaining a risk prediction model integrating an attention mechanism.

[0050] Input the physiological anomaly index, operation deviation degree, and environmental hazard coefficient into the risk prediction model of the integrated attention mechanism, and the dynamic risk value of the next period can be output.

[0051] 104. Correct the dynamic risk value of the next period according to the risk time-effect decay factor, and match the corrected dynamic risk value with the target risk level in the preset risk level.

[0052] Among them, the risk time-effect decay factor is determined by combining the dynamic risk value of the historical period and the real-time state parameters of the vehicle.

[0053] Specifically, the detailed execution process of determining the risk time-effect decay factor is as follows: conduct a temporal trend quantification analysis on the dynamic risk value of the historical period, conduct a quantitative evaluation on the real-time state parameters of the vehicle, and obtain the risk time-effect decay factor; correct the dynamic risk value based on the risk time-effect decay factor to obtain the corrected dynamic risk value. In this step, the dynamic risk values of each historical period are collected in advance. Use statistical methods or machine learning models to analyze the trend changes of historical dynamic risk values, including but not limited to the moving average method, exponential smoothing method, and survival analysis, etc. Among them, the moving average method: calculate the moving average values under different window sizes to smooth data fluctuations and identify long-term trends; the exponential smoothing method: assign higher weights to recent data to capture short-term trend changes; survival analysis: such as Kaplan-Meier estimation or Cox proportional hazards model, which is used to evaluate the "survival" probability of specific risk factors over time, and then infer the attenuation speed of their influence.

[0054] Determine the risk time-effect decay factor according to the above trend analysis results. , which represents the attenuation degree of the historical risk value over time. The specific expression is:

[0055] where is the basic exponential decay, the core decay term, which ensures that the historical dynamic risk value decreases over time, is the basic attenuation amount, and t is the time difference from the current time; is the trend enhancement factor, which dynamically adjusts the decay speed according to the trend (rising or falling) of the historical risk value. When the risk trend is rising (such as the risk value increasing for consecutive days), the enhancement factor may be positive, making the decay factor decrease (i.e., the decay becomes slower). When the risk trend is falling, the enhancement factor is negative, making the attenuation factor increase (i.e., accelerating the decay), is the trend sensitivity coefficient, and Trend(t) is the trend slope calculated by moving average or linear regression; is the covariate adjustment factor, which adjusts the decay speed according to the historical volatility (such as standard deviation) of the historical dynamic risk value; g(S) is the weighted correction term of the vehicle real-time state parameters, and n represents the number of vehicle real-time state parameters participating in the weighted correction, is the weight coefficient of the i-th vehicle real-time state parameter, is normalized to [0,1], that is, the original vehicle real-time state parameter is transformed into the range of [0,1], enabling parameters of different types and dimensions to participate in calculation and analysis on a unified scale, making the parameter value centered on 0.5 (such as when the battery power =0.3, (0.3 - 0.5) = -0.2, reducing the decay factor).

[0056] In the process of calculating the risk time-effect decay factor, exponential decay, trend analysis, and covariate adjustment are integrated. It combines the classical method of time decay (exponential decay) and the cutting-edge technology of trend analysis (such as the Cox model). Through the trend enhancement factor, it ensures sensitivity to the rising trend of risks, avoids ignoring potential risks due to excessive attenuation, can more flexibly reflect the attenuation law of historical dynamic risk values, and effectively improves the accuracy and real-time performance of risk prediction.

[0057] Combining the original dynamic risk value with the risk time-effect decay factor, the corrected dynamic risk value is obtained. The specific expression is: ;

[0058] where represents the original dynamic risk value, The corresponding risk time-effect decay factor, and g(S) is the weighted correction term of the vehicle's real-time state parameters.

[0059] It should be noted that in order to ensure that the correction process meets expectations and no obvious deviation is introduced. The above expression can be applied to the dynamic risk values of each historical period to generate a new sequence of corrected dynamic risk values, and a comparison chart of the dynamic risk values before and after correction can be visualized, so as to intuitively compare the changes before and after correction.

[0060] Correspondingly, a corresponding dynamic risk value interval can be set in advance for each risk level in the preset risk levels. The preset risk levels include low risk, medium risk, and high risk. When the corrected dynamic risk value is obtained, it can be matched with the risk value interval corresponding to the above-set preset risk levels, so as to determine the risk level of the next period, that is, the target risk level.

[0061] Through the above detailed implementation methods, by calculating the weighted correction term of the vehicle's real-time state parameters and introducing the risk time-effect decay factor, the corrected dynamic risk value takes into account both the continuous influence of the historical trend and the attenuation effect of the real-time state. The dynamic risk value correction logic better meets the requirements of "historical trend analysis + real-time state perception", improving the accuracy and scenario adaptability of risk level matching. Then, matching the target risk level according to the corrected dynamic risk value can significantly improve the accuracy, real-time performance, and robustness of risk assessment.

[0062] 105. Use the risk-privilege linkage rule to determine the dynamic privilege adjustment strategy corresponding to the target risk level.

[0063] Among them, the dynamic privilege adjustment strategy includes the hierarchical authorization threshold of function privileges and the real-time response rule.

[0064] In this step, a mapping relationship between each risk level and the predefined vehicle permissions is constructed, that is, the risk-permission linkage rule. Specifically, this risk-permission linkage rule can combine the permissions of all controllable functions in the vehicle as a permission database, and set a hierarchical authorization threshold and a real-time response rule for each risk level in the permission database. Among them, the hierarchical authorization threshold specifically refers to the allowed / limited function range and parameter threshold. Generally, it is divided into three levels: low risk, medium risk, and high risk, and the threshold can be dynamically adjusted based on safety indicators, traffic regulations, user habits, etc. For example, low risk: automatic lane change is allowed (distance to the following vehicle > 50 meters), cruise speed ≤ 120 km / h; medium risk: only adjacent lane change is allowed (distance to the following vehicle > 100 meters), cruise speed ≤ 100 km / h; high risk: lane change is prohibited, cruise speed ≤ 80 km / h. The real-time response rule includes three elements: response condition, execution action, and parameter limit. For example, condition: the camera / radar detects that the distance between the pedestrian and the vehicle itself < 20 meters and is at an unprotected intersection; action: immediately trigger automatic braking and disable the sound output of the entertainment system; parameter: braking deceleration ≥ 6 m / s², response time ≤ 0.3 seconds until the pedestrian leaves the dangerous area.

[0065] After determining the target risk level, through this risk-permission linkage rule, the dynamic permission adjustment strategy corresponding to the target risk level can be quickly determined.

[0066] 106. Based on the driving scenario where the vehicle is located, perform real-time verification on the dynamic permission adjustment strategy, generate a list of controllable functions with final authorization, and perform permission authorization operations in the next cycle to enable users to perform safety control.

[0067] In this step, through the permissions corresponding to the dynamic permission adjustment strategy determined in step 105, the corresponding controllable functions can be determined. In order to make the dynamic permission adjustment strategy more compatible with the driving scenario where the vehicle is located, adapt to the continuous changes in the long-term driving scenario, improve the safety redundancy and scenario adaptability of vehicle permission control, and thus ensure the collaborative safety and control flexibility of vehicle driving, real-time verification can be performed on the dynamic permission adjustment strategy through the driving scenario where the vehicle is located, so as to obtain a list of controllable functions with final authorization.

[0068] Specifically, the detailed execution process of performing real-time verification on the dynamic permission adjustment strategy based on the driving scenario where the vehicle is located and generating a list of controllable functions with final authorization is as follows: determine the driving scenario according to the road type, traffic environment, and road surface conditions where the vehicle is located; verify the hierarchical authorization threshold and the real-time response rule respectively based on the key safety indicators preset for the driving scenario to obtain the verification result; if the verification result is passed, generate a list of controllable functions according to the dynamic permission adjustment strategy; if the verification result is not passed, perform adaptive correction on the dynamic permission adjustment strategy and generate a list of controllable functions according to the corrected dynamic permission adjustment strategy.

[0069] In this step, through GPS / high-precision maps: locate the current road type (highway / urban road / tunnel), identify lane lines, traffic signs, traffic lights, pedestrians and vehicles through cameras, detect the density of surrounding obstacles, road boundaries and terrain changes with lidar (LiDAR), measure the distance and relative speed to the vehicle ahead with radar, so as to determine the traffic environment, and detect the weather (rain / snow / haze), road surface slipperiness and visibility through environmental sensors. Use a convolutional neural network (CNN) to extract features and classify the above data according to driving scenarios. For example: Highway scenario: vehicle speed > 80 km / h, clear lane lines, low traffic density. Urban road scenario: vehicle speed < 50 km / h, dense traffic lights, many pedestrians / non-motor vehicles. Tunnel scenario: weak GPS signal, low visibility, low speed limit. Extreme weather scenario: rainfall > 5 mm / h, visibility < 100 m.

[0070] Exemplarily, define key indicators for each driving scenario, as shown in Table 1: Table 1

[0071] Compare the hierarchical authorization threshold and the real-time response rule with the quantization criteria of their corresponding scenarios respectively against the key safety indicators to determine whether there are conflicts. If not, it means the verification passes, that is, the scenario rules and the dynamic permission policy have no conflicts and meet the safety indicators. At this time, generate a list of controllable functions according to the permissions corresponding to the dynamic permission adjustment policy. If there are conflicts, it means the verification fails, that is, the scenario rules are more restrictive or the dynamic policy violates the safety indicators. At this time, if the scenario rules require more strict restrictions (such as prohibiting autonomous driving in tunnels), the dynamic permission policy can be adjusted. If the safety indicators are not met (such as vehicle speed > 120 km / h), the real-time response rule can be modified, that is, trigger an emergency speed limit instruction, and generate a list of controllable functions according to the permissions corresponding to the corrected dynamic permission adjustment policy. After determining the list of controllable functions, permission control instructions can be sent to the ECU through the CAN bus, so as to retain the controllable functions for users to use, while other functions are disabled. In addition, the current target risk level and the list of available functions composed of controllable functions are displayed on the in-vehicle display screen, so that users can know the current situation of the vehicle by viewing the in-vehicle display screen, and voice broadcast can also be combined to remind users. For example, if the permissions are restricted, a voice prompt can be given: "Due to high risk, please take over driving immediately."

[0072] Based on the above Figure 1As can be seen from the implementation method, a method for controlling the permission security of an artificial intelligence vehicle provided by this application. The technical solution provided by this application synchronously collects physiological monitoring data, driving behavior data, and environmental perception data in the current cycle, calculates the physiological abnormality index, operation deviation degree, and environmental danger coefficient respectively, breaks through the limitations of single-dimensional risk assessment, realizes the all-factor risk coupling analysis of "human-vehicle-environment", avoids misjudgment caused by isolated judgment in traditional solutions, improves the comprehensiveness and accuracy of subsequent risk prediction, inputs the physiological abnormality index, operation deviation degree, and environmental danger coefficient in the current cycle into a risk prediction model that dynamically assigns global attention weights to the physiological abnormality index, operation deviation degree, and environmental danger coefficient according to the risk evolution relationship in the historical cycle, outputs the dynamic risk value in the next cycle, can capture the risk evolution trend, realizes risk pre-perception, enables permission adjustment to be ahead of the risk outbreak, corrects the dynamic risk value based on the risk time-effect decay factor, and matches the preset risk level after correction, considering the characteristic that the impact of risk events decays over time, making the matching of risk levels more accurate, generates a dynamic permission adjustment strategy through risk-permission linkage rules, and conducts real-time verification of the dynamic permission adjustment strategy based on the driving scenario where the vehicle is located, generates a list of manipulable functions with final authorization, can ensure the fit between the list of manipulable functions and the driving scenario, thus breaking the rigid limitations of static permission rules, realizing elastic expansion and contraction of permissions, while ensuring core safety functions, avoiding excessive restriction of reasonable operations by drivers, balancing safety and driving freedom, and continuously updating the "human-vehicle-environment" data, dynamic risk value, and list of manipulable functions in units of cycles, forming a closed loop of "perception-prediction-decision-execution", can adapt to the continuous changes in long-duration driving scenarios, thereby realizing real-time perception of multi-dimensional risks and flexibly adjusting vehicle function permissions, improving the safety redundancy and scenario adaptability of vehicle permission control, and further ensuring the collaborative safety and control flexibility of vehicle driving.

[0073] Furthermore, the preferred embodiment of this application is a detailed description of the process of controlling the permission security of an artificial intelligence vehicle on the basis of the above Figure 1 . The specific steps are as Figure 2 shown, including: 201. Obtain the physiological abnormality index, operation deviation degree, and environmental danger coefficient in the historical cycle.

[0074] In this step, historical driving data for the past 1 or 3 years is collected. The historical driving data specifically includes physiological monitoring data, driving behavior data, and environmental perception data for the historical period. Combining the calculation processes of the physiological anomaly index, operation deviation degree, and environmental danger coefficient for the current period in steps 101 - 103, the corresponding physiological anomaly index, operation deviation degree, and environmental danger coefficient for each historical period are calculated respectively based on the physiological monitoring data, driving behavior data, and environmental perception data for the historical period.

[0075] Perform spatio - temporal alignment and outlier processing on the above three types of data, and store them in the database, ensuring that each record contains the following fields: timestamp (accurate to seconds), physiological anomaly index, operation deviation degree, and environmental danger coefficient. This is for subsequent steps to call.

[0076] 202. Take the physiological anomaly index, operation deviation degree, and environmental danger coefficient as training data, and mark the risk level of the training data according to the real - risk event records to obtain training samples.

[0077] Among them, the real - risk event records include event type, occurrence timestamp, and severity.

[0078] In this step, real - risk event records such as actual accidents, congestions, and constructions can be obtained in advance from accident reports, traffic management departments, or insurance companies. Each record should include event type (such as accident, congestion), occurrence timestamp, severity (such as minor, medium, severe), etc.

[0079] Specifically, the detailed execution process of marking the risk level of the training data according to the real - risk event records to obtain training samples is as follows: Obtain the dynamic time window corresponding to different event types; according to the dynamic time window, match and associate the event type with the training data according to the occurrence timestamp; if the event type matched and associated in the training data is a single event, mark the risk value of the training data based on the severity corresponding to the single event; if the event type matched and associated in the training data is a composite event, mark the risk value of the training data according to the highest severity in the composite event.

[0080] It should be noted that the durations of the impacts of different types of risk events on driving behaviors and risk factors are different. For example, in the case of an accident: usually within a few minutes before and after the accident, significant changes will occur in the driver's behaviors (such as sudden braking and sharp steering wheel turning) and physiological states (such as rapid heartbeat); for congestion: it may have an impact for up to more than ten minutes or even dozens of minutes, because the vehicle needs to gradually decelerate and stop, and then start again; for construction: its impact may last even longer, because it not only changes the normal driving route but also may cause changes in traffic flow velocity. Therefore, considering the timeliness differences of different event types, different dynamic time windows can be set in advance according to the event types to better capture these differences, that is, to more accurately associate the historical physiological abnormality index, historical operation deviation degree, and historical environmental hazard coefficient with the actually occurring risk events, thereby improving the accuracy of the risk prediction model. Exemplarily, for an accident: the data within 5 minutes before and after the event occurrence is considered relevant; for congestion: the data within 15 minutes before and after the event occurrence is considered relevant; for construction: the data within 30 minutes before and after the event occurrence is considered relevant.

[0081] After obtaining the dynamic time windows corresponding to different time types, for each piece of training data (i.e., the historical physiological abnormality index, historical operation deviation degree, and historical environmental hazard coefficient at each time point), find all event records within its corresponding dynamic time window. If there is a single event, directly use the severity of this event as the risk label, and if there is a compound event, select the most severe event among them as the risk label. Exemplarily, low risk (such as no event or minor event); medium risk (such as general congestion); high risk (such as major accident or severe congestion).

[0082] In addition, in order to further improve the accuracy of the association between training data and event types, the dynamic time window also allows for precisely matching the physiological abnormality index, operation deviation degree, and environmental hazard coefficient of the corresponding historical period based on the occurrence location of the event. Thereby excluding those data points that are irrelevant to specific risk events, reducing unnecessary interference, which helps to improve the quality of model training and enables it to focus more on truly meaningful information.

[0083] 203. Construct an LSTM-Transformer hybrid network model as a risk prediction model integrating an attention mechanism.

[0084] Among them, the LSTM layer is used to capture the risk evolution relationship between the historical physiological abnormality index, historical operation deviation degree, and historical environmental hazard coefficient and the risk level, and the Transformer layer is used to calculate the global attention weights of the physiological abnormality index, historical operation deviation degree, and historical environmental hazard coefficient according to the risk evolution relationship.

[0085] In this step, the LSTM layer captures the risk evolution relationship. The LSTM (Long Short-Term Memory network) is used to model long-term dependencies in time series and capture the evolution law of physiological, operational, and environmental data over time. By inputting the multivariate time series data composed of three types of indicators (physiological anomaly index, operation deviation degree, environmental hazard coefficient) in the historical period, the hidden state at each time step can be output, representing the risk evolution characteristics of the current period. The Transformer layer calculates the global attention weights. The self-attention mechanism of the Transformer dynamically calculates the weights of each time step, giving higher attention to key periods (such as abnormal behaviors before high-risk events). By inputting the sequence of hidden states output by the LSTM layer, the following can be output: the weighted context vector, representing the global risk characteristics. Among them, the LSTM parameters: the number of hidden layer units: 64 - 256 (adjusted according to data complexity); bidirectional LSTM: enhances the model's ability to capture context features. Transformer parameters: the number of multi-head attention heads: 8 - 16; positional encoding: uses sine / cosine functions to embed time step information. The output layer can be for classification tasks: using the Softmax activation function to output the probability distribution (such as P(low)=0.1, P(medium)=0.5, P(high)=0.7); regression tasks: output the continuous risk value (such as the dynamic risk index between 0 and 1). The loss function of this risk prediction model is . Among them, = 0.7, which is used to ensure classification accuracy; = 0.3, which is used to ensure temporal continuity.

[0086] 204. Use the training samples to train the risk prediction model with the fused attention mechanism.

[0087] This step combines the description of the training process of the risk prediction model in the above method step 103. The same content will not be repeated here.

[0088] 205. Obtain the physiological monitoring data, driving behavior data of the user, and the environmental perception data of the vehicle in the current period.

[0089] This step combines the description of the above method step 101. The same content will not be repeated here.

[0090] 206. Perform spatio-temporal alignment and noise filtering processing on the physiological monitoring data, driving behavior data, and environmental perception data, and calculate the physiological anomaly index, operation deviation degree, and environmental hazard coefficient of the current period respectively.

[0091] This step combines the description of the above method step 102. The same content will not be repeated here.

[0092] 207. Input the physiological abnormality index, operation deviation degree, and environmental risk coefficient of the current cycle into the risk prediction model with a fusion attention mechanism to obtain the dynamic risk value of the next cycle.

[0093] This step is combined with the description of the above method step 103, and the same content will not be repeated here.

[0094] 208. Modify the dynamic risk value of the next cycle according to the risk time decay factor, and match the modified dynamic risk value to the target risk level in the preset risk levels.

[0095] This step is combined with the description of the above method step 104, and the same content will not be repeated here.

[0096] 209. Use the risk-privilege linkage rule to determine the dynamic privilege adjustment strategy corresponding to the target risk level.

[0097] This step is combined with the description of the above method step 105, and the same content will not be repeated here.

[0098] 210. Based on the driving scenario where the vehicle is located, perform real-time verification on the dynamic privilege adjustment strategy, generate a list of finally authorized controllable functions, and perform privilege authorization operations in the next cycle for the user to perform safety control.

[0099] This step is combined with the description of the above method step 106, and the same content will not be repeated here.

[0100] Furthermore, in order to further ensure the precise privilege adaptation between the dynamic risk level and the vehicle functions, thereby enhancing the collaborative safety and control flexibility of vehicle driving. On the basis of steps 201-210, it further includes: monitoring the user's satisfaction with the use of the list of controllable functions in the next cycle, and counting the risk event occurrence rate of the vehicle in the next cycle; calculating the safety fitness degree between the dynamic privilege adjustment strategy and the target risk level according to the satisfaction with the use and the risk event occurrence rate; if the safety fitness degree is lower than the preset threshold, trigger an update prompt message for the risk-privilege linkage rule.

[0101] In this step, the user's satisfaction can be determined through direct feedback, behavioral data, and sentiment analysis. Among them, direct feedback can collect the user's ratings on controllable functions (such as 1-5 star ratings) and review texts (such as "smooth operation", "response delay"); behavioral data can analyze the user's usage frequency, stay duration, number of incorrect operations, etc. of functions through logs; sentiment analysis can use natural language processing (NLP) technology to conduct sentiment polarity analysis on user comments (such as positive / neutral / negative), and extract negative feedback keywords (such as "lag", "insufficient permissions"). Specifically, a scoring system can be set up. This scoring system includes the quantitative scores corresponding to the indicators involved in the above three methods, calculates the scores of each method separately, and then performs weighted processing with their respective preset weights to calculate the satisfaction score. The range of this satisfaction score is 0-1.

[0102] The incidence rate of vehicle risk events can record high-risk events (such as operation errors, environmental hazard triggers, etc.) occurring in the next cycle through sensors, logs, or user reports, or can also combine the dynamic risk value correction results in steps 203-209 to count the occurrence frequency of actual risk events. Specifically, taking the current cycle as the benchmark, count the number and types of risk events (such as accidents, congestions, operation errors) occurring in the next cycle, so as to calculate the incidence rate of risk events.

[0103] After obtaining the user satisfaction and the incidence rate of risk events, the safety fit can be calculated by combining their respective preset importance factors. A fit threshold (such as 0.8) is preset in advance. Compare the safety fit with this fit threshold. If the safety fit is equal to or higher than the preset threshold, it means that the risk-permission linkage rule at this time does not need to be adjusted, and the collaborative safety and control flexibility of vehicle driving can be ensured. If the safety fit is lower than the preset threshold, it means that the risk-permission linkage rule needs to be adjusted to further improve the collaborative safety and control flexibility of vehicle driving. At this time, an update prompt message for the risk-permission linkage rule is triggered. This update prompt message includes: Suggestion: Automatically generate suggestions for the risk-permission linkage rule (such as "adjust the permission of the high-risk level from 'allowed' to'restricted'"), provide an attribution analysis of the incidence rate of risk events (such as "the operation error rate has increased by 20%"), etc. The notification method can be to display an alarm on the in-vehicle display screen or push an alarm through email, text message, etc., so that users can update the risk-permission linkage rule in time to ensure the collaborative safety and control flexibility of vehicle driving.

[0104] Furthermore, as for the above Figure 1 - Figure 2For the implementation of the method embodiments shown, embodiments of the present application provide a permission security control device for an intelligent vehicle. This device is used to predict the comprehensive risk value in the next period through multi-modal data of physiology, behavior, and environment, and achieve precise permission adaptation of vehicle functions based on the dynamic risk level, thereby enhancing the collaborative safety and control flexibility of vehicle driving. The embodiments of this device correspond to the foregoing method embodiments. For the convenience of reading, the details in the foregoing method embodiments will not be elaborated one by one in this embodiment. However, it should be clear that the device in this embodiment can correspondingly implement all the contents in the foregoing method embodiments. Specifically, as Figure 3 shown, the device includes: A first acquisition unit 31, configured to acquire the physiological monitoring data, driving behavior data of the user, and environmental perception data of the vehicle in the current period; A first calculation unit 32, configured to perform spatio-temporal alignment and noise filtering processing on the physiological monitoring data, the driving behavior data, and the environmental perception data, and calculate the physiological abnormality index, operation deviation degree, and environmental danger coefficient in the current period respectively; A prediction unit 33, configured to input the physiological abnormality index, operation deviation degree, and environmental danger coefficient in the current period into a risk prediction model with a fusion attention mechanism to obtain the dynamic risk value in the next period. The attention mechanism is used to dynamically allocate the global attention weights of the physiological abnormality index, the operation deviation degree, and the environmental danger coefficient according to the risk evolution relationship in the historical period; A processing unit 34, configured to correct the dynamic risk value in the next period according to the risk time decay factor, and match the corrected dynamic risk value to the target risk level in the preset risk level. The risk time decay factor is determined by combining the dynamic risk value in the historical period and the real-time state parameters of the vehicle; A determination unit 35, configured to determine the dynamic permission adjustment strategy corresponding to the target risk level by using the risk-permission linkage rule. The dynamic permission adjustment strategy includes the hierarchical authorization threshold of function permissions and the real-time response rule; An authorization unit 36, configured to perform real-time verification on the dynamic permission adjustment strategy based on the driving scenario where the vehicle is located, generate a list of finally authorized controllable functions, and perform permission authorization operations in the next period for the user to perform safety control.

[0105] Furthermore, as Figure 4 shown, the first acquisition unit 301 includes: A first collection module 3011, configured to collect the breathing frequency, micro-tremor amplitude, sitting posture stability, and the proportion of the duration of hand contact with the steering wheel of the user in the current period to obtain the physiological monitoring data of the user in the current period; The second acquisition module 3012 is configured to acquire the throttle pedal pressure gradient, brake pedal pressure gradient, steering angular velocity, and vehicle acceleration within the current cycle, so as to obtain the driving behavior data of the user within the current cycle; The first acquisition module 3013 is configured to acquire obstacle information, lane line status information, traffic flow information, and traffic event information in the radar point cloud data of the vehicle surrounding environment within the current cycle, so as to obtain the environmental perception data of the vehicle within the current cycle.

[0106] Further, as Figure 4 shown, the calculation unit 302 includes: The preprocessing module 3021 is configured to perform spatio-temporal alignment and noise filtering processing on the physiological monitoring data, the driving behavior data, and the environmental perception data, so as to obtain the processed physiological monitoring data, driving behavior data, and environmental perception data; The first determination module 3022 is configured to calculate a respiratory rate anomaly score, a micro-tremor amplitude anomaly score, a sitting posture stability anomaly score, and a contact duration anomaly score respectively based on the processed physiological monitoring data, and determine the physiological anomaly index according to the respiratory rate anomaly score, the micro-tremor amplitude anomaly score, the sitting posture stability anomaly score, and the contact duration anomaly score; The second determination module 3023 is configured to calculate a throttle pressure gradient deviation degree, a brake pressure gradient deviation degree, a steering angular velocity deviation degree, and a vehicle acceleration deviation degree respectively based on the processed driving behavior data, and determine the operation deviation degree according to the throttle pressure gradient deviation degree, the brake pressure gradient deviation degree, the steering angular velocity deviation degree, and the vehicle acceleration deviation degree; The third determination module 3024 is configured to calculate an obstacle risk degree, a lane line integrity degree, a traffic flow risk degree, and an event urgency degree respectively based on the processed environmental perception data, and determine the environmental risk coefficient according to the obstacle risk degree, the lane line integrity degree, the traffic flow risk degree, and the event urgency degree.

[0107] Further, as Figure 4 shown, the device further includes: The second acquisition unit 307 is configured to acquire the physiological anomaly index, operation deviation degree, and environmental risk coefficient of the historical cycle before inputting the physiological anomaly index, operation deviation degree, and environmental risk coefficient of the current cycle into the risk prediction model of the fusion attention mechanism to obtain the dynamic risk value of the next cycle; The marking unit 308 is configured to use the physiological anomaly index, operation deviation degree, and environmental risk coefficient of the historical cycle as training data, and perform risk value marking on the training data according to the real risk event record to obtain training samples; A construction unit 309 is used to construct an LSTM-Transformer hybrid network model as the risk prediction model of the fusion attention mechanism. The LSTM layer is used to capture the risk evolution relationship between the physiological anomaly index, operation deviation degree, and environmental hazard coefficient in the historical period and the risk level. The Transformer layer is used to calculate the global attention weights of the physiological anomaly index, operation deviation degree, and environmental hazard coefficient according to the risk evolution relationship. A training unit 310 is used to train the risk prediction model of the fusion attention mechanism by using the training samples.

[0108] Further, as Figure 4 shown, the real risk event record includes event type, occurrence event timestamp, and severity; the marking unit 308 includes: A second acquisition module 3081 is used to acquire dynamic time windows corresponding to different event types. An association module 3082 is used to match and associate the event type with the training data according to the dynamic time window and the occurrence event timestamp. A marking module 3083 is used to mark the risk value of the training data based on the severity corresponding to the single event if the event type matched and associated in the training data is a single event. The marking module 3083 is further used to mark the risk value of the training data according to the highest severity in the compound event if the event type matched and associated in the training data is a compound event.

[0109] Further, the authorization unit 306 includes: A fourth determination module 3061 is used to determine the driving scenario according to the road type, traffic environment, and road surface conditions where the vehicle is located. A verification module 3062 is used to verify the hierarchical authorization threshold and the real-time response rule respectively based on the key safety indicators preset for the driving scenario, and obtain a verification result. A generation module 3063 is used to generate the list of controllable functions according to the dynamic permission adjustment strategy if the verification result is passed. The generation module 3063 is used to adaptively correct the dynamic permission adjustment strategy and generate the list of controllable functions according to the corrected dynamic permission adjustment strategy if the verification result is not passed.

[0110] Further, as Figure 4 shown, the device further includes: The monitoring and statistics unit 311 is configured to monitor the user's satisfaction with the use of the manipulable function list in the next cycle and count the incidence rate of risk events of the vehicle in the next cycle; The second calculation unit 312 is configured to calculate the safety compliance between the dynamic permission adjustment policy and the target risk level according to the use satisfaction and the incidence rate of risk events; The trigger unit 313 is configured to trigger an update prompt message for the risk-permission linkage rule if the safety compliance is lower than a preset threshold.

[0111] Furthermore, an embodiment of the present application further provides a storage medium for storing a computer program, wherein the computer program controls the device where the storage medium is located to execute the above-mentioned Figure 1 - Figure 2 permission security control method of the artificial intelligence vehicle described above.

[0112] Furthermore, an embodiment of the present application further provides a processor for running a program, wherein the program executes the permission security control method of the artificial intelligence vehicle described above when running. Figure 1 - Figure 2 In the above embodiments, the descriptions of the respective embodiments have their own emphases. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0113] It can be understood that the relevant features in the above methods and devices can be referred to each other. In addition, the "first", "second", etc. in the above embodiments are used to distinguish the respective embodiments, and do not represent the advantages and disadvantages of the respective embodiments.

[0114] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0115]

[0116] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings provided herein. The structure required to construct such a system will be apparent from the above description. In addition, the present application is not directed to any specific programming language. It should be understood that the content of the present application described herein can be implemented using various programming languages, and the description of the specific language above is to disclose the best mode of the present application.

[0117] ​In addition, the memory may include non-permanent memory in the form of computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flashRAM), and the memory includes at least one memory chip.

[0118] Those skilled in the art will appreciate that the embodiments of the present application may be provided as a method, system, or computer program product. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0119] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0120] These computer program instructions can also be stored in a computer-readable memory capable of guiding the computer or other programmable data processing devices to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, such that a series of operation steps are performed on the computer or other programmable devices to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or combinations of blocks.

[0122] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0123] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0124] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0125] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0126] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0127] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for controlling the permission security of an artificial intelligence vehicle, characterized in that, The method includes: Obtaining the physiological monitoring data, driving behavior data of the user, and environmental perception data of the vehicle within the current cycle; Performing spatio-temporal alignment and noise filtering processing on the physiological monitoring data, the driving behavior data, and the environmental perception data, and respectively calculating the physiological abnormality index, operation deviation degree, and environmental danger coefficient of the current cycle; Inputting the physiological abnormality index, operation deviation degree, and environmental danger coefficient of the current cycle into a risk prediction model with a fusion attention mechanism to obtain the dynamic risk value of the next cycle, where the attention mechanism is used to dynamically allocate the global attention weights of the physiological abnormality index, the operation deviation degree, and the environmental danger coefficient according to the risk evolution relationship of the historical cycle; Correcting the dynamic risk value of the next cycle according to the risk time decay factor, and matching the corrected dynamic risk value to the target risk level in the preset risk levels, where the risk time decay factor is determined by combining the dynamic risk value of the historical cycle and the real-time state parameters of the vehicle; Determining the dynamic permission adjustment strategy corresponding to the target risk level by using the risk-permission linkage rule, where the dynamic permission adjustment strategy includes the hierarchical authorization threshold of the function permission and the real-time response rule; Performing real-time verification on the dynamic permission adjustment strategy based on the driving scenario where the vehicle is located, generating a list of finally authorized controllable functions, and performing the permission authorization operation in the next cycle for the user to perform safety control.

2. The method according to claim 1, wherein Obtaining the physiological monitoring data, driving behavior data of the user, and environmental perception data of the vehicle within the current cycle, including: Collecting the breathing frequency, micro-tremor amplitude, sitting posture stability, and the proportion of the duration of hand contact with the steering wheel of the user within the current cycle to obtain the physiological monitoring data of the user within the current cycle; Collecting the throttle pedal pressure gradient, brake pedal pressure gradient, steering angular velocity, and vehicle acceleration within the current cycle to obtain the driving behavior data of the user within the current cycle; Obtaining the obstacle information, lane line status information, traffic flow information, and traffic event information in the radar point cloud data of the vehicle surrounding environment within the current cycle to obtain the environmental perception data of the vehicle within the current cycle.

3. The method according to claim 2, wherein Performing spatio-temporal alignment and noise filtering processing on the physiological monitoring data, the driving behavior data, and the environmental perception data, and respectively calculating the physiological abnormality index, operation deviation degree, and environmental danger coefficient of the current cycle, including: Performing spatio-temporal alignment and noise filtering processing on the physiological monitoring data, the driving behavior data, and the environmental perception data to obtain the processed physiological monitoring data, driving behavior data, and environmental perception data; Based on the processed physiological monitoring data, respectively calculating the breathing frequency abnormality score, micro-tremor amplitude abnormality score, sitting posture stability abnormality score, and contact duration abnormality score, and determining the physiological abnormality index according to the breathing frequency abnormality score, the micro-tremor amplitude abnormality score, the sitting posture stability abnormality score, and the contact duration abnormality score; Calculate the throttle pressure gradient deviation, brake pressure gradient deviation, steering angular velocity deviation, and vehicle acceleration deviation based on the processed driving behavior data respectively, and determine the operation deviation according to the throttle pressure gradient deviation, the brake pressure gradient deviation, the steering angular velocity deviation, and the vehicle acceleration deviation; Calculate the obstacle risk level, lane line integrity, traffic flow risk level, and event urgency based on the processed environmental perception data respectively, and determine the environmental risk coefficient according to the obstacle risk level, the lane line integrity, the traffic flow risk level, and the event urgency.

4. The method according to claim 1, wherein Before inputting the physiological abnormality index, operation deviation, and environmental risk coefficient of the current cycle into the risk prediction model of the fusion attention mechanism to obtain the dynamic risk value of the next cycle, the method further includes: Obtain the physiological abnormality index, operation deviation, and environmental risk coefficient of the historical cycle; use the physiological abnormality index, operation deviation, and environmental risk coefficient of the historical cycle as training data, and perform risk value marking on the training data according to the real risk event record to obtain training samples; Construct an LSTM-Transformer hybrid network model as the risk prediction model of the fusion attention mechanism. The LSTM layer is used to capture the risk evolution relationship between the physiological abnormality index, operation deviation, and environmental risk coefficient of the historical cycle and the risk level. The Transformer layer is used to calculate the global attention weights of the physiological abnormality index, the operation deviation, and the environmental risk coefficient according to the risk evolution relationship; use the training samples to train the risk prediction model of the fusion attention mechanism.

5. The method according to claim 4, wherein The real risk event record includes event type, occurrence event timestamp, and severity; Performing risk value marking on the training data according to the real risk event record to obtain training samples, including: Obtain the dynamic time window corresponding to different event types; According to the dynamic time window, match and associate the event type with the training data according to the occurrence event timestamp; If the event type matched and associated in the training data is a single event, perform risk value marking on the training data based on the severity corresponding to the single event; If the event type matched and associated in the training data is a composite event, perform risk value marking on the training data according to the highest severity in the composite event.

6. The method according to claim 1, characterized in that Perform real-time verification on the dynamic permission adjustment strategy based on the driving scenario of the vehicle to generate a list of finally authorized controllable functions, including: Determine the driving scenario according to the road type, traffic environment, and road surface conditions of the vehicle; Verify the hierarchical authorization threshold and the real-time response rule respectively based on the key safety indicators preset for the driving scenario to obtain a verification result; If the verification result is passed, generate the list of controllable functions according to the dynamic permission adjustment strategy; If the verification result fails, adaptively correct the dynamic permission adjustment policy, and generate the list of controllable functions according to the corrected dynamic permission adjustment policy.

7. The method according to claim 1, characterized in that, The method further includes: Monitoring the user's satisfaction with the use of the list of controllable functions in the next cycle, and counting the incidence rate of risk events of the vehicle in the next cycle; Calculating the safety compliance degree between the dynamic permission adjustment policy and the target risk level according to the use satisfaction and the incidence rate of risk events; If the safety compliance degree is lower than a preset threshold, trigger an update prompt message for the risk-permission linkage rule.

8. A permission security control device for an artificial intelligence vehicle, characterized in that, Applied to the permission security control method of the artificial intelligence vehicle according to any one of claims 1-7 above, the device includes: A first acquisition unit for acquiring the physiological monitoring data, driving behavior data of the user and the environmental perception data of the vehicle in the current cycle; A first calculation unit for performing spatio-temporal alignment and noise filtering processing on the physiological monitoring data, the driving behavior data and the environmental perception data, and respectively calculating the physiological abnormality index, operation deviation degree and environmental danger coefficient of the current cycle; A prediction unit for inputting the physiological abnormality index, operation deviation degree and environmental danger coefficient of the current cycle into a risk prediction model with a fusion attention mechanism to obtain the dynamic risk value of the next cycle, and the attention mechanism is used to dynamically allocate the global attention weights of the physiological abnormality index, the operation deviation degree and the environmental danger coefficient according to the risk evolution relationship of the historical cycle; A processing unit for correcting the dynamic risk value of the next cycle according to the risk time decay factor, and matching the corrected dynamic risk value with the target risk level in the preset risk level, and the risk time decay factor is determined by combining the dynamic risk value of the historical cycle and the real-time state parameters of the vehicle; A determination unit for determining the dynamic permission adjustment policy corresponding to the target risk level by using the risk-permission linkage rule, and the dynamic permission adjustment policy includes the hierarchical authorization threshold of function permissions and the real-time response rule; An authorization unit for performing real-time verification on the dynamic permission adjustment policy based on the driving scenario where the vehicle is located, generating a list of finally authorized controllable functions, and performing permission authorization operations in the next cycle for the user to perform safety control.

9. A storage medium, characterized in that, The storage medium includes a stored program, wherein when the program runs, it controls the device where the storage medium is located to execute the permission security control method of the artificial intelligence vehicle according to any one of claims 1 to 7.

10. A processor, characterized in that, The processor is used to run the program, wherein when the program runs, it executes the permission security control method of the artificial intelligence vehicle according to any one of claims 1 to 7.

Images