A vehicle driving risk state prediction method

Abstract

The present application relates to the field of vehicle safety, and discloses a vehicle driving risk state prediction method, which comprises the following steps: collecting running state data, steering input data and relative environment data in the process of vehicle driving in real time, constructing a multi-dimensional time series data set of vehicle driving state; performing data point outlier analysis based on the data set to determine the deviation degree of the current vehicle running state, so as to determine whether the current vehicle is in an outlier state; then constructing a vehicle driving state trajectory and calculating the trajectory geometric curvature, analyzing the dynamic change trend to identify whether the vehicle driving has a sharp mutation; then constructing a multi-dimensional driving state space, analyzing the trajectory shape aggregation degree to predict whether the vehicle is facing a loss of control risk, and monitoring the instability trajectory evolution rate to calculate the risk urgency index; finally, defining the safety feasible region boundary according to the risk level, evaluating the stability margin and calculating the state correction vector, and outputting the risk state evaluation result, so as to realize the accurate prediction and quantitative correction of the risk.

Description

Technical Field

[0001] This invention relates to the field of vehicle safety technology, and more specifically to a method for predicting the risk status of vehicle driving. Background Technology

[0002] With the rapid development of intelligent connected vehicle technology, vehicle driving safety research has become a core issue in the fields of advanced driver assistance systems and autonomous driving. Ensuring driving safety through real-time monitoring and risk assessment of vehicle operating status is particularly important. Existing vehicle driving risk prediction technologies mainly fall into two categories: methods based on vehicle dynamics models and data-driven methods. The former typically constructs a stability domain boundary in the state space based on the vehicle's force balance equations or phase plane theory to determine whether the vehicle is unstable. The latter utilizes time-series data collected by multiple sensors to analyze and identify the degree of deviation or trajectory evolution characteristics of the vehicle's driving state by constructing a reference model of normal driving behavior or using machine learning algorithms.

[0003] However, the aforementioned existing technologies still have certain limitations in practical applications. First, prediction methods based on vehicle dynamics models often rely on pre-defined fixed stability domain boundaries, making it difficult to dynamically adjust boundary conditions according to changes in the vehicle's real-time driving risk level. This results in insufficient adaptability and accuracy of the prediction models under complex and changing driving conditions. Second, existing data-driven methods mostly focus on determining the static deviation of the vehicle's current state from historical data, lacking in-depth mining and fusion analysis of the geometrical abrupt changes in the state trajectory evolution process and the rate of instability trend evolution over time. This makes it difficult to accurately quantify the urgency of risk occurrence in the early stages. Furthermore, existing risk assessment results usually only remain at the level of qualitative risk level indication, lacking quantitative calculation of the specific state adjustment direction and amplitude required for the vehicle to recover to a safe and feasible domain. This limits the guiding value of risk prediction results for proactive intervention by drivers or vehicle control systems. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for predicting the risk state of vehicle driving, so as to solve the problems existing in the background art.

[0005] This invention provides the following technical solution: a method for predicting vehicle driving risk status, comprising:

[0006] S1: Construct a multi-dimensional time series dataset of vehicle driving status by collecting real-time operating status data, control input data, and relative environmental data during vehicle driving.

[0007] S2: Based on the constructed multidimensional time series dataset, perform outlier analysis on the data points to determine the degree of deviation of the current vehicle operating status from the historical normal driving distribution characteristics, so as to determine whether the current vehicle is in an outlier state.

[0008] S3: Based on the determination that the vehicle is currently in an outlier state, construct the vehicle's driving trajectory and calculate the trajectory's geometric curvature. Analyze the dynamic change trend of the vehicle's driving state and determine the trajectory evolution abrupt change characteristic value to identify whether the vehicle's driving trend has undergone a drastic change.

[0009] S4: Based on the mutation identification results, construct a multi-dimensional driving state space, analyze the clustering degree of the state trajectory morphology, and calculate the state instability characteristic value to predict whether the vehicle faces the risk of losing control.

[0010] S5: Based on the predicted risk of loss of control of the vehicle, monitor and analyze the evolution rate of the unstable trajectory pattern, calculate the driving risk urgency index, and determine the driving risk level of the vehicle.

[0011] S6: Based on the determined vehicle driving risk level, define the boundary of the safe and feasible region in the multi-dimensional driving state space, calculate the distance between the current trajectory point and the boundary of the safe and feasible region, and evaluate the stability margin of the vehicle driving state.

[0012] S7: Based on the stability margin of the vehicle's driving state, analyze and calculate the state correction vector required for the vehicle to recover to the safe and feasible region, and output the multi-dimensional driving risk state assessment results.

[0013] Preferably, step S1 utilizes an onboard sensing system to collect real-time operating status data, control input data, and relative environmental data during vehicle operation. The operating status data includes the vehicle's longitudinal speed, yaw rate, roll rate, and center of gravity sideslip angle. The control input data includes the steering wheel angle, accelerator pedal opening, and brake line pressure. The relative environmental data includes the relative distance to the vehicle in front, relative speed, and road surface adhesion coefficient. The operating status data, control input data, and relative environmental data are filtered, denoised, and normalized, and further time-synchronized and serialized to construct a multi-dimensional time-series dataset of the vehicle's driving status.

[0014] Preferably, step S2 constructs a reference model of the distribution characteristics of historical normal driving of vehicles based on a multidimensional time series dataset. The reference model is represented by a multidimensional mean vector and a covariance matrix. A sliding time window is used to extract the multidimensional time series data within the current time window. Based on the extracted data, the multidimensional mean vector and covariance matrix at the current moment are calculated in real time and the parameters in the original reference model are replaced to obtain a dynamically updated reference model.

[0015] Based on the updated multidimensional mean vector and covariance matrix, the Mahalanobis distance of the vehicle driving status data points at the current moment is calculated. The Mahalanobis distance is used as the degree of deviation of the current vehicle driving status from the historical normal driving distribution characteristics. The Mahalanobis distance is compared with a preset outlier determination threshold. If the Mahalanobis distance is greater than the preset outlier determination threshold, the current vehicle is determined to be in an outlier state, and vehicle driving status trajectory analysis is triggered. Otherwise, real-time monitoring continues.

[0016] Preferably, in response to the determination that the vehicle is currently in an outlier state, step S3 extracts key state variables from a multidimensional time-series dataset. These key state variables include the vehicle's longitudinal velocity, yaw rate, and steering wheel angle. Based on these key state variables, a multidimensional phase space trajectory is reconstructed. Based on the coordinates of the trajectory points in the multidimensional phase space trajectory, the trajectory velocity vector and trajectory acceleration vector are calculated. The trajectory geometric curvature is then solved according to the vector operation relationship between the trajectory velocity vector and the trajectory acceleration vector. A time-series difference operation is performed on the trajectory geometric curvature to obtain the rate of curvature change. This rate of curvature change is used as a feature value of trajectory evolution abrupt change to reflect the dynamic trend of the vehicle's driving state.

[0017] The trajectory evolution abrupt change feature value is compared with the preset abrupt change judgment threshold. If the trajectory evolution abrupt change feature value is greater than the preset abrupt change judgment threshold, it is determined that the vehicle driving trend has undergone a drastic change. Otherwise, it is determined that the vehicle driving trend has not undergone a drastic change, and monitoring continues.

[0018] Preferably, in response to the recognition result of a drastic change in the vehicle's driving trend, step S4 extracts the centroid sideslip angle and yaw rate from the multidimensional time series dataset as coordinate axes to construct a two-dimensional phase plane of the vehicle's driving state; selects trajectory points within a preset time period before the current moment in this two-dimensional phase plane, calculates the geometric center position of all trajectory points; then calculates the Euclidean distance between each trajectory point and the geometric center position, and takes the average of all Euclidean distances to obtain the trajectory distribution radius; the trajectory distribution radius is used as a state instability feature value to reflect the degree of clustering of the state trajectory shape, wherein a larger trajectory distribution radius represents a lower degree of clustering and a higher risk of instability;

[0019] The unstable state characteristic value is compared with the preset unstable state judgment threshold. If the unstable state characteristic value is greater than the preset unstable state judgment threshold, the vehicle is predicted to face the risk of loss of control; otherwise, the vehicle is predicted to be in a critical stable state.

[0020] Preferably, in response to the prediction result that the vehicle faces the risk of losing control, S5 calculates the difference between the state instability characteristic value at the current moment and the previous moment, and calculates the ratio of the absolute value of the difference to the sampling time interval to obtain the instability evolution speed, so as to represent the instantaneous severity of the evolution of the instability trajectory; then calculates the difference between the instability evolution speed at the current moment and the previous moment, and calculates the ratio of the absolute value of the difference to the sampling time interval to obtain the instability evolution acceleration, so as to predict the trend intensity of the evolution of the instability trajectory.

[0021] The evolution direction is determined by the sign of the difference in the state instability characteristic values. If the difference is positive, it is determined to be a divergent direction; if the difference is negative, it is determined to be a convergent direction. When the evolution direction is divergent, a dynamic weighting coefficient is calculated based on the magnitude of the instability evolution acceleration. This dynamic weighting coefficient is the sum of the preset benchmark weight value and the absolute value of the instability evolution acceleration. The product of the dynamic weighting coefficient and the instability evolution velocity is then calculated, and the result of this product is used as the driving risk urgency index. When the evolution direction is convergent, the driving risk urgency index is set to zero or a preset safety value.

[0022] The driving risk urgency index is matched with a preset multi-level risk threshold range, and the vehicle driving risk level is output according to the numerical range of the matching result.

[0023] Preferably, step S6 matches the corresponding boundary correction coefficient according to the determined vehicle driving risk level, and uses the boundary correction coefficient to dynamically correct the vehicle dynamic stability boundary conditions, so as to define the boundary of the safe and feasible domain in the multi-dimensional driving state space.

[0024] Obtain the coordinate position of the vehicle's trajectory point in the multi-dimensional driving state space at the current moment; calculate the shortest Euclidean distance between the current trajectory point and the boundary of the safe and feasible domain; define the shortest Euclidean distance as the stability margin, which refers to the remaining safe distance of the vehicle driving dynamics system from the instability critical point.

[0025] Preferably, step S7 determines the target recovery point within the safe feasible region along the shortest path based on the current trajectory point and the boundary of the safe feasible region in the multidimensional driving state space; calculates the multidimensional coordinate difference between the target recovery point and the current trajectory point to generate a state correction vector, which represents the direction and magnitude of the state parameter adjustment required for the vehicle to recover to the safe feasible region; integrates the vehicle driving risk level, stability margin value, and state correction vector to construct a multidimensional driving risk state assessment result; the multidimensional driving risk state assessment result includes risk level prompt information, stability margin value, and state correction vector, wherein the risk level prompt information is used to drive the vehicle-mounted display terminal to provide visual warnings.

[0026] The technical effects and advantages of this invention are as follows:

[0027] (1) By matching the corresponding boundary correction coefficient according to the determined vehicle driving risk level to dynamically define the boundary of the safe and feasible domain, the defect of the traditional dynamic model relying on fixed boundaries is overcome, and the boundary of the safe and feasible domain is adaptively corrected according to the real-time risk level, which effectively improves the adaptability and accuracy of the prediction model under complex and variable driving conditions.

[0028] (2) By constructing the vehicle driving state trajectory and calculating the trajectory geometric curvature to identify the evolution mutation characteristics, and combining the analysis of the evolution rate of the unstable trajectory shape to calculate the driving risk urgency index, the deep mining and spatiotemporal fusion analysis of the geometric shape mutation characteristics and the evolution rate of the time dimension during the evolution of the state trajectory is realized, which can accurately quantify the urgency of the risk occurrence in the early stage.

[0029] (3) By analyzing and calculating the state correction vector required for the vehicle to recover to the safe and feasible domain based on the stability margin of the vehicle's driving state, it can provide a multi-dimensional driving risk state assessment result that includes risk level, stability margin value and state correction vector. It realizes the leap from qualitative risk warning to quantitative state correction and provides the driver or vehicle control system with operable active intervention guidance. Attached Figure Description

[0030] Figure 1 This is a diagram illustrating the method steps of the present invention.

[0031] Figure 2 This is a system structure block diagram of the present invention. Detailed Implementation

[0032] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The vehicle driving risk state prediction method involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] like Figure 1 The embodiment shown provides a method for predicting vehicle driving risk status, including:

[0034] S1: By collecting real-time data on the vehicle's operating status, control input, and relative environment during the driving process, a multi-dimensional time series dataset of the vehicle's driving status is constructed.

[0035] In this embodiment, step S1 utilizes an onboard sensing system to collect real-time operating status data, control input data, and relative environmental data during vehicle operation. The operating status data includes the vehicle's longitudinal speed, yaw rate, roll rate, and center of gravity sideslip angle. The control input data includes the steering wheel angle, accelerator pedal opening, and brake line pressure. The relative environmental data includes the relative distance to the vehicle in front, relative speed, and road surface adhesion coefficient. The operating status data, control input data, and relative environmental data are filtered, denoised, and normalized, and further time-synchronized and serialized to construct a multi-dimensional time-series dataset of the vehicle's driving status.

[0036] Specifically, the vehicle-mounted sensing system includes an inertial measurement unit (IMU) mounted on the vehicle chassis, an angle sensor mounted on the steering column, displacement sensors mounted on the accelerator and brake pedals, a millimeter-wave radar mounted on the front of the vehicle, and wheel speed sensors mounted near the tires. The IMU collects real-time data on the vehicle's longitudinal velocity, yaw rate, roll rate, and center of gravity sideslip angle at a sampling frequency of 50 Hz. The angle sensor collects real-time data on the steering wheel angle at a sampling frequency of 50 Hz. The displacement sensor collects real-time data on the accelerator pedal opening and brake line pressure at a sampling frequency of 50 Hz. The millimeter-wave radar collects real-time data on the relative distance and relative speed to the vehicle in front at a sampling frequency of 20 Hz. The wheel speed sensors, combined with a road surface recognition algorithm, estimate the road adhesion coefficient in real-time at a sampling frequency of 10 Hz. The raw data from each channel is filtered and denoised using a fourth-order Butterworth low-pass filter with a cutoff frequency set to 10 Hz to eliminate noise. High-frequency noise interference; the filtered channel data is normalized using the min-max normalization method, which involves subtracting the minimum value of the historical data for that channel from the value of each data point and then dividing by the difference between the maximum and minimum values ​​of the historical data for that channel, so that all data are uniformly mapped to the interval between 0 and 1; for the problem of inconsistent timestamps caused by different hardware sampling frequencies of different sensors, a linear interpolation method is used to uniformly resample all channel data to a unified time reference of 50 Hz to achieve time synchronization; the time-synchronized channel data is arranged in chronological order to form a data vector containing 10 dimensions for each moment, including longitudinal velocity, yaw rate, roll rate, center of gravity sideslip angle, steering wheel angle, accelerator pedal opening, brake line pressure, relative distance to the vehicle in front, relative speed to the vehicle in front, and road adhesion coefficient. The data vectors of all moments are concatenated in chronological order to form a multi-dimensional time series dataset of vehicle driving status.

[0037] S2: Based on the constructed multidimensional time series dataset, perform outlier analysis on the data points to determine the degree of deviation of the current vehicle operating status from the historical normal driving distribution characteristics, so as to determine whether the current vehicle is in an outlier state.

[0038] In this embodiment, S2 constructs a reference model of the distribution characteristics of the vehicle's historical normal driving based on a multidimensional time series dataset. The reference model is represented by a multidimensional mean vector and a covariance matrix. A sliding time window is used to extract the multidimensional time series data within the current time window. Based on the extracted data, the multidimensional mean vector and covariance matrix at the current moment are calculated in real time and the parameters in the original reference model are replaced to obtain a dynamically updated reference model.

[0039] Based on the updated multidimensional mean vector and covariance matrix, the Mahalanobis distance of the vehicle driving status data points at the current moment is calculated. The Mahalanobis distance is used as the degree of deviation of the current vehicle driving status from the historical normal driving distribution characteristics. The Mahalanobis distance is compared with a preset outlier determination threshold. If the Mahalanobis distance is greater than the preset outlier determination threshold, the current vehicle is determined to be in an outlier state, and vehicle driving status trajectory analysis is triggered. Otherwise, real-time monitoring continues.

[0040] It should be specifically noted that when constructing the reference model for the distribution characteristics of historical normal driving of vehicles, the 10-dimensional data vector at each moment in the multi-dimensional time series dataset is used as a sample to calculate the mean vector of all samples under historical normal driving conditions. Covariance Matrix , where the mean vector It is a 10-dimensional column vector, where the i-th element is the arithmetic mean of all historical normal samples in the i-th channel, and the covariance matrix is... 10 A symmetric matrix of size 10, where the element in the i-th row and i-th column is the covariance of all historical normal samples from the i-th channel and the i-th channel. A sliding time window is used to extract multidimensional time series data within the current time window. The sliding window length is set to 200 sampling points, corresponding to a time span of 4 seconds, and the sliding step size is set to 1 sampling point (0.02 seconds). Each sliding step extracts the latest 200 data points within the window as the current window sample set. The mean vector at the current time is calculated in real time based on the current window sample set. With covariance matrix The calculation method is the same as that of the historical reference model. and ∑ replaces the original reference model and A dynamically updated reference model is obtained; based on the updated mean vector Covariance Matrix The formula for calculating the Mahalanobis distance L of data point X at the current time is L √(X, ...). transpose multiplied by Multiply the inverse matrix by (X minus) The outlier detection threshold is not a fixed constant, but is dynamically generated based on the statistical characteristics of the historical Mahalanobis distance sequence within the current window. Specifically, it involves extracting all 200 Mahalanobis distance values ​​within the current window and calculating their mean. and standard deviation The outlier detection threshold Set as Add 3 times ,Right now equal Add 3 When the data follows an approximately normal distribution, this threshold corresponds to a 99.7% confidence interval; if the Mahalanobis distance L at the current time is greater than the threshold... If the current vehicle is out of the group, the current data point is marked as an anomaly, and a trigger signal is sent to the downstream trajectory analysis module. Upon receiving the trigger signal, the downstream module immediately initiates the trajectory construction and curvature calculation process of S3. If L is less than or equal to the threshold... If the trajectory analysis is not triggered, the system continues to slide the window forward in steps of 0.02 seconds and repeat the outlier analysis.

[0041] S3: Based on the determination that the vehicle is currently in an outlier state, construct the vehicle's driving trajectory and calculate the trajectory's geometric curvature. Analyze the dynamic change trend of the vehicle's driving state and determine the trajectory evolution abrupt change characteristic value to identify whether the vehicle's driving trend has undergone a drastic change.

[0042] In this embodiment, S3 responds to the determination result that the current vehicle is in an outlier state, extracts key state variables from the multidimensional time series dataset, including the vehicle's longitudinal velocity, yaw rate, and steering wheel angle, and reconstructs the multidimensional phase space trajectory based on the key state variables; based on the coordinates of the trajectory points of the multidimensional phase space trajectory, the trajectory velocity vector and trajectory acceleration vector are calculated, and the trajectory geometric curvature is solved according to the vector operation relationship between the trajectory velocity vector and the trajectory acceleration vector; the trajectory geometric curvature is subjected to time series difference operation to obtain the rate of curvature change, and the rate of curvature change is used as the trajectory evolution abrupt change feature value to reflect the dynamic change trend of the vehicle's driving state;

[0043] The trajectory evolution abrupt change feature value is compared with the preset abrupt change judgment threshold. If the trajectory evolution abrupt change feature value is greater than the preset abrupt change judgment threshold, it is determined that the vehicle driving trend has undergone a drastic change. Otherwise, it is determined that the vehicle driving trend has not undergone a drastic change, and monitoring continues.

[0044] It should be specifically noted that, in response to the determination that the current vehicle is in an outlier state, the vehicle's longitudinal velocity v and yaw rate are extracted from the multi-dimensional time series dataset. and steering wheel angle There are three key state variables, and a 3D phase space is constructed using these three variables as coordinate axes. The phase space trajectory is reconstructed using a time-delay embedding method, specifically: selecting data points from N consecutive sampling times before the current time t, and then embedding them with a time delay. As an interval, the 3D vector at each sampling time is delayed. Arranged sequentially, they form a phase space vector with an embedding dimension of m, where N is 200. Five sampling points are taken, i.e., 0.1 seconds, and m is set to 3. Each phase space point is composed of concatenated 3D vectors from the current time, the time 0.1 seconds ago, and the time 0.2 seconds ago, forming a total of 9-dimensional embedding vectors. All embedding vectors arranged in chronological order constitute the multidimensional phase space trajectory. After extending the 3D key state variables to a 9-dimensional phase space through time-delay embedding, the calculation of the trajectory geometric curvature needs to adapt to the characteristics of high-dimensional space. Specifically, the trajectory velocity vector B and trajectory acceleration vector A are calculated based on two adjacent embedding vectors in the phase space trajectory, and the trajectory geometric curvature is calculated according to the high-dimensional space curve curvature formula. Specifically, the area of ​​the parallelogram spanned by the magnitudes and inner product of the velocity and acceleration vectors is calculated. Taking the numerator as the numerator and the cube of the magnitude of the velocity vector as the denominator, the curvature is obtained by dividing the numerator by the denominator. Perform a first-order forward difference operation on the curvature values ​​at consecutive time points, that is, subtract the curvature of the previous time point from the curvature of the current time point and divide by the time interval of 0.02 seconds to obtain the rate of change of curvature. Based on the rate of change of curvature, the curvature acceleration is further calculated, i.e., the rate of change of curvature is subjected to another first-order forward difference. The curvature acceleration is used as the second-order abrupt change characteristic of trajectory evolution to capture the accelerating trend of curvature change. The abrupt change characteristic value of trajectory evolution is defined as the comprehensive abrupt change characteristic value Q, where Q is derived from the rate of change of curvature. The result is obtained by weighting and fusing the curvature acceleration with weights of 0.6 and 0.4, respectively; that is, Q equals 0.6 multiplied by 0.4. The absolute value is added to 0.4 multiplied by the absolute value of curvature acceleration; the mutation determination threshold does not use a fixed constant, but is calculated based on the comprehensive mutation feature value sequence within the last 50 sampling times, i.e., within 1 second, to calculate a dynamic threshold. Specifically, the mean Q and standard deviation of all comprehensive mutation feature values ​​within the last 1 second are calculated. , set the mutation detection threshold Set to Q plus 2.5 times If the comprehensive mutation characteristic value Q is greater than the threshold If the vehicle's driving trend changes drastically, it is determined that a drastic change has occurred; otherwise, it is determined that no drastic change has occurred, and monitoring continues at a cycle of 0.02 seconds.

[0045] S4: Based on the mutation identification results, construct a multi-dimensional driving state space, analyze the clustering degree of the state trajectory morphology, and calculate the state instability feature value to predict whether the vehicle faces the risk of losing control.

[0046] In this embodiment, S4 responds to the recognition result of a drastic change in the vehicle's driving trend by extracting the centroid sideslip angle and yaw rate from the multidimensional time series dataset as coordinate axes to construct a two-dimensional phase plane of the vehicle's driving state; selects trajectory points within a preset time period before the current moment in this two-dimensional phase plane and calculates the geometric center position of all trajectory points; then calculates the Euclidean distance between each trajectory point and the geometric center position, and takes the average of all Euclidean distances to obtain the trajectory distribution radius; the trajectory distribution radius is used as a state instability feature value to reflect the degree of clustering of the state trajectory shape, wherein a larger trajectory distribution radius represents a lower degree of clustering and a higher risk of instability;

[0047] The unstable state characteristic value is compared with the preset unstable state judgment threshold. If the unstable state characteristic value is greater than the preset unstable state judgment threshold, the vehicle is predicted to face the risk of loss of control; otherwise, the vehicle is predicted to be in a critical stable state.

[0048] It should be specifically noted that, in response to the identification results of drastic changes in vehicle driving trends, the centroid sideslip angle is extracted from the multi-dimensional time-series dataset. With yaw rate Two state variables are used as coordinate axes to construct a two-dimensional phase plane of the vehicle's driving state; trajectory points within this two-dimensional phase plane that are 3 consecutive seconds (150 sampling times) before the current time t are selected and denoted as point set P. Each trajectory point The coordinates are the centroid sideslip angle at the i-th sampling time. With yaw rate The two-dimensional vector is composed of points P; the geometric center C of the point set P is calculated, where the x-coordinate of C is the arithmetic mean of the centroid sideslip angles of the 150 trajectory points. =1 divided by 150 multiplied by the sum of the values ​​from i equal to 1 to 150 The ordinate of C is the arithmetic mean of the yaw angular velocities of 150 trajectory points. =1 divided by 150 multiplied by the sum of the values ​​from i equal to 1 to 150 ; Calculate each trajectory point Euclidean distance from the geometric center C , equal to the square root ( reduce squared plus ( reduce The square of 150 Euclidean distances is used to calculate the arithmetic mean of the distances from i to i. The radius R of the trajectory distribution is obtained by dividing 1 by 150 and summing the distances from i to i. Using only the trajectory distribution radius R as an instability criterion is insufficient because R only reflects the overall dispersion of the trajectory and cannot characterize its morphological and directional features. For example, when the trajectory exhibits a highly anisotropic elliptical distribution, although the R value may be normal, the vehicle may have entered a specific risk state such as drifting or skidding. Therefore, it is necessary to further calculate the covariance matrix of the trajectory point set. , It is a 2x2 symmetric matrix, whose main diagonal elements are the variances of the centroid sideslip angle sequence and the yaw rate sequence, respectively, and whose secondary diagonal elements are the covariances of the two; for the covariance matrix... Eigenvalue decomposition yields two eigenvalues. and ,in Greater than or equal to Calculate the anisotropy ratio H equal to Divide by A larger H indicates a more severe stretching of the trajectory points along a certain direction, meaning a more unstable shape. Based on anisotropy analysis, a trajectory complexity index is further introduced, and box counting is used to estimate the fractal dimension of the trajectory. Specifically, the trajectory coordinates in the two-dimensional phase plane are first normalized to a unit square region using the minimum-maximum method, and then the coordinates are calculated using a side length of... A square grid covers the area; count the number of grids that intersect with the trajectory. ,Change Four scales were selected: 0.5, 0.25, 0.125, and 0.0625, and the corresponding values ​​were recorded. ,by For the horizontal axis, A linear fit is performed on the vertical axis, and the slope of the fitted line is the fractal dimension. , The closer the value is to 2, the denser and more chaotic the trajectory filling, indicating a stronger tendency towards instability; the trajectory distribution radius R, anisotropy ratio H, and fractal dimension are also considered. After normalization, the three values ​​are weighted and fused with weights of 0.5, 0.3, and 0.2 to obtain the comprehensive instability characteristic value S; that is, S equals 0.5 multiplied by the normalized value of R plus 0.3 multiplied by the normalized value of H plus 0.2 multiplied by the normalized value of H. The normalized value is calculated by subtracting the minimum value of each indicator from the values ​​within the last 100 sampling times and then dividing by the difference between the maximum and minimum values. The instability threshold is not a fixed constant but is calculated based on a dynamic threshold derived from the sequence of comprehensive instability characteristic values ​​over the last 100 sampling times (i.e., within 2 seconds). Specifically, this involves calculating the mean of all comprehensive instability characteristic values ​​within the last 2 seconds. and standard deviation The instability determination threshold Set as Add 2 times If the comprehensive instability characteristic value S is greater than the threshold If the system predicts that the vehicle is at risk of losing control, it sends a loss-of-control warning signal to the downstream risk level determination module; otherwise, if the system predicts that the vehicle is in a critical stable state, it continues to monitor and repeat the S4 analysis process at a cycle of 0.02 seconds.

[0049] S5: Based on the predicted risk of loss of control of the vehicle, monitor and analyze the evolution rate of the unstable trajectory pattern, calculate the driving risk urgency index, and determine the driving risk level of the vehicle.

[0050] In this embodiment, S5 responds to the prediction result of the vehicle facing the risk of loss of control by calculating the difference between the current moment and the previous moment's state instability characteristic value, and calculates the ratio of the absolute value of the difference to the sampling time interval to obtain the instability evolution speed, so as to represent the instantaneous severity of the instability trajectory morphology evolution; then calculates the difference between the current moment and the previous moment's instability evolution speed, and calculates the ratio of the absolute value of the difference to the sampling time interval to obtain the instability evolution acceleration, so as to predict the trend intensity of the instability trajectory morphology evolution;

[0051] The evolution direction is determined by the sign of the difference in the state instability characteristic values. If the difference is positive, it is determined to be a divergent direction; if the difference is negative, it is determined to be a convergent direction. When the evolution direction is divergent, a dynamic weighting coefficient is calculated based on the magnitude of the instability evolution acceleration. This dynamic weighting coefficient is the sum of the preset benchmark weight value and the absolute value of the instability evolution acceleration. The product of the dynamic weighting coefficient and the instability evolution velocity is then calculated, and the result of this product is used as the driving risk urgency index. When the evolution direction is convergent, the driving risk urgency index is set to zero or a preset safety value.

[0052] The driving risk urgency index is matched with a preset multi-level risk threshold range, and the vehicle driving risk level is output according to the numerical range of the matching result.

[0053] It should be noted that the sampling time interval for the prediction of the risk of the vehicle losing control is 0.02 seconds. Perform the following calculation: Denote the state instability characteristic value at the current time t as... The previous moment State instability characteristic value Calculate the difference between the two. equal minus ,right Take the absolute value ,Will Divide by That is, 0.02 seconds, to obtain the rate of instability evolution. ,Right now equal Divide by 0.02, The unit is per second; a larger value indicates a more drastic evolution of the unstable trajectory. The rate of instability evolution at the previous moment is denoted as . Calculate the current instability evolution and The difference equal minus ,right Take the absolute value ,Will Divide by That is, 0.02 seconds, to obtain the acceleration of the unstable evolution. ,Right now equal Divide by 0.02, The unit is squared per second; a dimensionless coefficient k is introduced (taken as the reciprocal of the reference time step, such as 50), based on the difference in state instability characteristic values. The positive or negative sign determines the direction of evolution, if A value greater than 0 indicates that the divergence direction, or the degree of instability, is intensifying. If the value is less than 0, it is determined that the convergence direction, i.e., the degree of instability, is easing. If the value is 0, the current state is maintained; when the evolution direction is divergent, the preset baseline weight value is used. Taking 1.0, the dynamic weighting coefficient W is calculated to be equal to... Add k multiplied by That is, W equals 1.0 plus Then calculate the driving risk urgency index U, which equals W multiplied by That is, U equals (1.0 plus) Multiply by When the evolution direction is convergent, to avoid the risk being erroneously eliminated due to small fluctuations at a single moment, a convergence protection mechanism is set: convergence protection only occurs when there are 5 consecutive sampling moments, i.e., within 0.1 seconds. The driving risk urgency index U is set to 0 only when all values ​​are less than 0. If any one of the five consecutive sampling times is less than 0, the risk urgency index U is set to 0. If the value is greater than or equal to 0, the calculation will still be performed according to the divergence direction, but... Multiplying by the attenuation factor of 0.3 before calculating U, U equals (1.0 plus...). ) multiplied by (0.3 multiplied by Based on the driving risk urgency index U, the cumulative risk exposure E is further calculated. E is an approximate time integral of U over nearly 200 sampling times, i.e., 4 seconds. Specifically, it is the sum of all U values ​​over nearly 4 seconds and then multiplied by . That is, 0.02 seconds, which means E equals 0.02 multiplied by the sum of the values ​​from i equal to 1 to 200. E reflects the cumulative risk level of a vehicle under sustained high-pressure conditions in the near future. The driving risk pressure index U and the cumulative risk exposure E are both normalized. The normalization method is to subtract the minimum value of each index over the past 200 sampling times and then divide by the difference between the maximum and minimum values ​​to obtain the result. and Then, by weighting the data with 0.7 and 0.3 respectively, a comprehensive risk index is obtained. ,Right now Equal to 0.7 multiplied by Add 0.3 multiplied by Set four risk threshold ranges: Level 1 security response Less than 0.2, Level 2 warning corresponds to A value greater than or equal to 0.2 and less than 0.5 corresponds to Level 3 hazard. A value greater than or equal to 0.5 and less than 0.8 corresponds to Level 4 runaway. Greater than or equal to 0.8; To prevent frequent jumps in risk levels between adjacent levels due to small fluctuations in U, a lag comparison mechanism is introduced: if the current comprehensive risk index is greater than or equal to 0.8; The transition from a lower level to a higher level requires... An upgrade is only confirmed when the threshold exceeds 1.1 times the lower limit of the target level (e.g., from level 1 to level 2, it needs to be...). If the current The change from a higher level to a lower level requires A downgrade is only confirmed when the price falls below 0.9 times the lower limit of the target level; based on the comprehensive risk index. The system matches the numerical range with the corresponding risk level and outputs the result. The system continues to perform monitoring and calculation repeatedly every 0.02 seconds.

[0054] S6: Based on the determined vehicle driving risk level, define the boundary of the safe and feasible region in the multi-dimensional driving state space, calculate the distance between the current trajectory point and the boundary of the safe and feasible region, and evaluate the stability margin of the vehicle driving state.

[0055] In this embodiment, step S6 matches the corresponding boundary correction coefficient according to the determined vehicle driving risk level, and uses the boundary correction coefficient to dynamically correct the vehicle dynamic stability boundary conditions in order to define the boundary of the safe and feasible domain in the multi-dimensional driving state space.

[0056] Obtain the coordinate position of the vehicle's trajectory point in the multi-dimensional driving state space at the current moment; calculate the shortest Euclidean distance between the current trajectory point and the boundary of the safe and feasible domain; define the shortest Euclidean distance as the stability margin, which refers to the remaining safe distance of the vehicle driving dynamics system from the instability critical point.

[0057] It should be noted that the corresponding boundary correction coefficient is matched based on the determined vehicle driving risk level. The risk level is divided into 4 levels, corresponding to The values ​​are 1.0, 0.85, 0.7, and 0.55 respectively, indicating that the higher the risk level, the better. A smaller value indicates that the safe and feasible region is compressed more; this is achieved using boundary correction coefficients. The vehicle dynamic stability boundary conditions are dynamically modified. The vehicle dynamic stability boundary conditions include three aspects: the first is the centroid sideslip angle. stability limit Take 8 degrees, which is 0.14 radians. The second value is the yaw rate. stability limit Take 0.8 radians per second, the third one is the lateral acceleration. stability limit Take 0.4g, which is 3.92 meters per second squared, and multiply the above three stability limits by... The corrected dynamic stability boundary is obtained, that is, the corrected centroid sideslip angle limit is: Multiply by 0.14 radians, the corrected yaw rate limit is: Multiply by 0.8 radians per second, the corrected lateral acceleration limit is Multiply by 3.92 m / s²; in the multi-dimensional driving state space, with the sideslip angle at the center of mass. yaw rate Lateral acceleration A three-dimensional state space is constructed using three variables as coordinate axes. The boundary of the safe and feasible region is enclosed by the three modified dynamic stability limits mentioned above, forming an ellipsoid centered at the origin in the three-dimensional space. The length of the semi-axis in the axial direction is Multiply by 0.14, in The length of the semi-axis in the axial direction is Multiply by 0.8, in The length of the semi-axis in the axial direction is Multiply by 3.92; obtain the coordinates of the vehicle's trajectory point in the 3D state space at the current time t. Its three components are the current centroid sideslip angle. Current yaw rate Current lateral acceleration ; Calculate the current trajectory point The shortest Euclidean distance between the ellipsoidal boundary of the safe feasible region and the boundary of the safe feasible region The specific calculation method is as follows: the equation of the ellipsoid is written as... For trajectory points Calculate the ratio u equal to the square root of the root. If u is greater than or equal to 1, then the trajectory point is outside the safe feasible region or on the boundary. A value of 0 indicates that there is no stability margin. If u is less than 1, the trajectory point is within the safe and feasible region. equal Multiply Multiply by 0.14 and Multiply by 0.8 and Multiply by 3.92 to the weighted harmonic mean of the three factors, where the weights of the harmonic mean are 0.5, 0.3, and 0.2 respectively. , , Three directions, A larger value indicates a more sufficient stability margin; only the shortest Euclidean distance is used. There is a lack of stability margin because The system cannot reflect the direction from which the trajectory points approach the boundary, therefore further calculation of the trajectory points is required. The direction vector of the ellipsoid center, i.e., the origin, is The angle between the line vectors connecting to the nearest boundary point , The closer the value is to 0 degrees, the more stable the trajectory point is, indicating that it moves directly away from the boundary along the shortest path. The closer to 90 degrees, the more likely the trajectory point is to slip along the tangent, indicating a high risk of instability, though the direction is dangerous. At the shortest Euclidean distance... Angle with direction Based on this, a boundary curvature correction factor is introduced, specifically: calculating the principal curvature of the ellipsoid at the nearest boundary point along the approach direction of the trajectory point. and second curvature principal curvature Let be the curvature value of the ellipsoid at that point along the direction of maximum curvature, and be the second degree of curvature. The curvature ratio is calculated as the curvature value perpendicular to the direction of maximum curvature. equal Divide by , A larger value indicates a sharper boundary in that direction, meaning even a slight deviation could lead to boundary breach; the shortest Euclidean distance... , direction angle cosine value curvature ratio After normalization, the three factors are weighted and fused with weights of 0.5, 0.25, and 0.25 to obtain the overall stability margin. ,Right now equals 0.5 multiplied by The normalized value plus 0.25 multiplied by The normalized value plus 0.25 multiplied by (1 divided by) The normalized value of ) is obtained by subtracting the minimum value of the indicator within approximately 100 sampling times (i.e., 2 seconds) from each indicator and then dividing by the difference between the maximum and minimum values; a stability margin safety criterion is set: if If the value is greater than 0.6, the vehicle is considered to be in a safe condition. If the value is greater than or equal to 0.3 and less than or equal to 0.6, the vehicle is considered to be in a critical state. If the value is less than 0.3, the vehicle is considered to be in a dangerous state; at the same time, margin change rate monitoring is introduced to calculate the comprehensive stability margin at the current moment. Compared to the previous moment The difference equal minus ,like If the value is less than 0 for three consecutive sampling times (0.06 seconds), a margin drop warning is triggered, indicating that the stability margin is rapidly shrinking and immediate intervention is required; the system continues to repeat the calculation and evaluation of S6 with a period of 0.02 seconds.

[0058] S7: Based on the stability margin of the vehicle's driving state, analyze and calculate the state correction vector required for the vehicle to recover to the safe and feasible region, and output the multi-dimensional driving risk state assessment results.

[0059] In this embodiment, step S7 determines the target recovery point within the safe feasible region along the shortest path based on the current trajectory point and the boundary of the safe feasible region in the multidimensional driving state space; calculates the multidimensional coordinate difference between the target recovery point and the current trajectory point to generate a state correction vector, which represents the direction and magnitude of the state parameter adjustment required for the vehicle to recover to the safe feasible region; integrates the vehicle driving risk level, stability margin value, and state correction vector to construct a multidimensional driving risk state assessment result; the multidimensional driving risk state assessment result includes risk level prompt information, stability margin value, and state correction vector, wherein the risk level prompt information is used to drive the vehicle display terminal to provide visual warnings.

[0060] It should be specifically noted that this refers to the trajectory point at the current time t in the multi-dimensional driving state space. To determine the boundary of the safe and feasible region, perform the following calculations: First, determine the trajectory points. Relative to the location of the safe and feasible region, if If u is less than or equal to 1 within the safe and feasible region, then the vehicle is determined to be in a stable state, and the state correction vector is determined accordingly. Setting it to zero vector indicates that no stability intervention is required; if Outside the safe and feasible region, i.e., u is greater than 1, then along The shortest path direction to the surface of the ellipsoid is selected as the safety margin compensation amount within the boundary, at a distance of depth from the boundary. The point is used as the target recovery point Safety margin compensation amount Equals 0.3 multiplied by the stability margin Multiply by the weighted arithmetic mean semi-axis length of the ellipsoid in that direction, i.e. Equal to 0.3 multiplied by Multiply by (0.5 multiplied by) Multiply by 0.14 and add 0.3 multiplied by Multiply by 0.8 and add 0.2 multiplied by Multiply by 3.92); Target recovery point The method for calculating coordinates is as follows: First calculate The unit direction vector Z pointing to the center O of the ellipsoid is negative. Divide by The modulus, calculation Approximate radial distance to the surface of the ellipsoid The total correction distance is equal to (u-1) multiplied by the weighted arithmetic mean semi-axis length. equal Plus Target recovery point equal Plus Multiply by Z; then calculate the state correction vector. equal minus , It is a 3D vector with three components as follows: equal minus , equal minus , equal minus , The model is This represents the total magnitude of the state correction. The direction represents the optimal correction direction; since directly using Mapping vehicle control commands may exceed the physical limits of the actuators, so a feasibility correction constraint is introduced: calculate the three components separately. , , Whether it exceeds their respective single-step correction range, the single-step correction range is set as follows: Not exceeding ±2 degrees, or ±0.035 radians. Not exceeding ±0.15 radians per second The correction energy should not exceed ±0.5 m / s². If any component exceeds this limit, it should be truncated to the corresponding boundary value, and the truncation flag should be recorded. Simultaneously, the remaining untruncated components should be renormalized to maintain the total correction energy constant. (In the state correction vector...) Based on this, further calculations are made to estimate the recovery time. , equal Divide by the corrected rate constant Corrected rate constant A value of 0.8 means that 0.8 units of state deviation can be eliminated per second. The unit is seconds. A recovery time greater than 5 seconds is classified as slow recovery and requires higher-level intervention; integrate vehicle driving risk levels. Overall stability margin and state correction vector The three components and the modulus and recovery time estimate By combining the truncation flag, a multi-dimensional driving risk status assessment result is constructed. This result is output in the form of a structured data packet containing 6 fields: the first field is the risk level. Values ​​1 to 4 correspond to safe, warning, danger, and out of control, respectively. The second field represents the stability margin. Accurate to 3 decimal places, the third field is the 3 components of the state correction vector. , , Accurate to 4 decimal places, the fourth field is the total correction value. Accurate to 3 decimal places, the 5th field is the estimated recovery time. Accurate to two decimal places, the sixth field is a truncation flag: 0 indicates no truncation, 1 indicates truncation. Risk level warning information is used to drive the vehicle-mounted display terminal for visual warnings. Specifically, risk levels are distinguished by color on the display terminal: green for safe, yellow for warning, orange for danger, and red for out of control. A status correction vector is also overlaid on the display terminal in the form of arrows. In the 2D projection plane, i.e. The arrow's direction and length on the plane are shown, with the starting point being the projected position of the current trajectory point and the ending point being the projected position of the target recovery point. The arrow color matches the risk level, and the arrow length is scaled proportionally to the screen display area. An estimated recovery time is also indicated next to the arrow. The system continues to repeat the calculation and output of S7 with a cycle of 0.02 seconds.

[0061] like Figure 2This embodiment provides an implementation system corresponding to a vehicle driving risk state prediction method, including a multi-dimensional driving data acquisition module, a driving state outlier detection module, a trajectory curvature mutation identification module, a trajectory clustering instability prediction module, an evolution rate urgency rating module, a safety domain stability margin assessment module, and a driving risk state result output module. The multi-dimensional driving data acquisition module is connected to the driving state outlier detection module, the driving state outlier detection module is connected to the trajectory curvature mutation identification module, the trajectory curvature mutation identification module is connected to the trajectory clustering instability prediction module, the trajectory clustering instability prediction module is connected to the evolution rate urgency rating module, the evolution rate urgency rating module is connected to the safety domain stability margin assessment module, and the safety domain stability margin assessment module is connected to the driving risk state result output module.

[0062] Multi-dimensional driving data acquisition module: Constructs a multi-dimensional time series dataset of vehicle driving status by collecting real-time operating status data, control input data, and relative environmental data during vehicle driving.

[0063] Driving Status Outlier Detection Module: Based on the constructed multidimensional time series dataset, outlier analysis of data points is performed to determine the degree of deviation of the current vehicle operating status from the historical normal driving distribution characteristics, so as to determine whether the current vehicle is in an outlier state.

[0064] The trajectory curvature abrupt change identification module: Based on the determination result that the current vehicle is in an outlier state, it constructs the vehicle's driving state trajectory, calculates the trajectory geometric curvature, analyzes the dynamic change trend of the vehicle's driving state, and determines the trajectory evolution abrupt change characteristic value to identify whether the vehicle's driving trend has undergone a drastic change.

[0065] Trajectory Clustering Instability Prediction Module: Based on the mutation identification results, a multi-dimensional driving state space is constructed, and the clustering degree of the state trajectory morphology is analyzed to calculate the state instability feature value in order to predict whether the vehicle is facing the risk of losing control.

[0066] Evolution Rate Urgency Rating Module: Based on the predicted risk of loss of control of the vehicle, it monitors and analyzes the evolution rate of the unstable trajectory pattern, calculates the driving risk urgency index, and determines the driving risk level of the vehicle.

[0067] Safety Domain Stability Margin Assessment Module: Based on the determined vehicle driving risk level, the module defines the boundary of the safe and feasible domain in the multi-dimensional driving state space, calculates the distance between the current trajectory point and the boundary of the safe and feasible domain, and assesses the stability margin of the vehicle driving state.

[0068] Driving risk state result output module: Based on the stability margin of the vehicle's driving state, analyze and calculate the state correction vector required for the vehicle to recover to the safe and feasible region, and output the multi-dimensional driving risk state assessment results.

[0069] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0070] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for predicting vehicle driving risk status, characterized in that, include: S1: Construct a multi-dimensional time series dataset of vehicle driving status by collecting real-time operating status data, control input data, and relative environmental data during vehicle driving. S2: Based on the constructed multidimensional time series dataset, perform outlier analysis on the data points to determine the degree of deviation of the current vehicle operating status from the historical normal driving distribution characteristics, so as to determine whether the current vehicle is in an outlier state. S3: Based on the determination that the vehicle is currently in an outlier state, construct the vehicle's driving trajectory and calculate the trajectory's geometric curvature. Analyze the dynamic change trend of the vehicle's driving state and determine the trajectory evolution abrupt change characteristic value to identify whether the vehicle's driving trend has undergone a drastic change. S4: Based on the mutation identification results, construct a multi-dimensional driving state space, analyze the clustering degree of the state trajectory morphology, and calculate the state instability characteristic value to predict whether the vehicle faces the risk of losing control. S5: Based on the predicted risk of loss of control of the vehicle, monitor and analyze the evolution rate of the unstable trajectory pattern, calculate the driving risk urgency index, and determine the driving risk level of the vehicle. S6: Based on the determined vehicle driving risk level, define the boundary of the safe and feasible region in the multi-dimensional driving state space, calculate the distance between the current trajectory point and the boundary of the safe and feasible region, and evaluate the stability margin of the vehicle driving state. S7: Based on the stability margin of the vehicle's driving state, analyze and calculate the state correction vector required for the vehicle to recover to the safe and feasible region, and output the multi-dimensional driving risk state assessment results.

2. The method for predicting vehicle driving risk status according to claim 1, characterized in that, S1 utilizes an onboard sensing system to collect real-time operational status data, control input data, and relative environmental data during vehicle operation. The operational status data includes the vehicle's longitudinal speed, yaw rate, roll rate, and center of gravity sideslip angle. The control input data includes steering wheel angle, accelerator pedal opening, and brake line pressure. The relative environmental data includes the relative distance to the vehicle in front, relative speed, and road surface adhesion coefficient. The operational status data, control input data, and relative environmental data are filtered, denoised, and normalized, and further time-synchronized and serialized to construct a multi-dimensional time-series dataset of the vehicle's driving status.

3. The method for predicting vehicle driving risk status according to claim 2, characterized in that, S2 constructs a reference model of the distribution characteristics of historical normal driving of vehicles based on a multidimensional time series dataset. The reference model is represented by a multidimensional mean vector and a covariance matrix. A sliding time window is used to extract the multidimensional time series data within the current time window. Based on the extracted data, the multidimensional mean vector and covariance matrix at the current moment are calculated in real time and the parameters in the original reference model are replaced to obtain a dynamically updated reference model. Based on the updated multidimensional mean vector and covariance matrix, the Mahalanobis distance of the vehicle driving state data points at the current moment is calculated, and the Mahalanobis distance is used as the degree of deviation of the current vehicle driving state from the historical normal driving distribution characteristics. The Mahalanobis distance is compared with a preset outlier threshold. If the Mahalanobis distance is greater than the preset outlier threshold, the vehicle is determined to be in an outlier state, and vehicle driving trajectory analysis is triggered. Otherwise, real-time monitoring continues.

4. The method for predicting vehicle driving risk status according to claim 3, characterized in that, In response to the determination that the current vehicle is in an outlier state, S3 extracts key state variables from the multidimensional time series dataset. The key state variables include the vehicle's longitudinal velocity, yaw rate, and steering wheel angle. Based on the key state variables, the multidimensional phase space trajectory is reconstructed. Based on the coordinates of trajectory points in the multi-dimensional phase space trajectory, the trajectory velocity vector and trajectory acceleration vector are calculated, and the trajectory geometric curvature is solved according to the vector operation relationship between the trajectory velocity vector and the trajectory acceleration vector. The trajectory geometric curvature is subjected to time series difference operation to obtain the curvature change rate, and the curvature change rate is used as the trajectory evolution abrupt change characteristic value to reflect the dynamic change trend of vehicle driving state. The trajectory evolution abrupt change feature value is compared with the preset abrupt change judgment threshold. If the trajectory evolution abrupt change feature value is greater than the preset abrupt change judgment threshold, it is determined that the vehicle driving trend has undergone a drastic change. Otherwise, it is determined that the vehicle driving trend has not undergone a drastic change, and monitoring continues.

5. The method for predicting vehicle driving risk status according to claim 4, characterized in that, S4 responds to the recognition result of a drastic change in vehicle driving trend, extracts the centroid sideslip angle and yaw rate from the multidimensional time series dataset as coordinate axes, and constructs a two-dimensional phase plane of vehicle driving state; selects trajectory points within a preset time period before the current moment in this two-dimensional phase plane, calculates the geometric center position of all trajectory points; then calculates the Euclidean distance between each trajectory point and the geometric center position, and takes the average of all Euclidean distances to obtain the trajectory distribution radius; the trajectory distribution radius is used as a state instability feature value to reflect the degree of clustering of the state trajectory shape, wherein the larger the trajectory distribution radius, the lower the clustering degree and the higher the risk of instability; The unstable state characteristic value is compared with the preset unstable state judgment threshold. If the unstable state characteristic value is greater than the preset unstable state judgment threshold, the vehicle is predicted to face the risk of loss of control; otherwise, the vehicle is predicted to be in a critical stable state.

6. The method for predicting vehicle driving risk status according to claim 5, characterized in that, S5 responds to the prediction result of the vehicle facing the risk of loss of control, calculates the difference between the state instability characteristic value at the current moment and the previous moment, and calculates the ratio of the absolute value of the difference to the sampling time interval to obtain the instability evolution speed, so as to represent the instantaneous severity of the evolution of the instability trajectory; then calculates the difference between the instability evolution speed at the current moment and the previous moment, and calculates the ratio of the absolute value of the difference to the sampling time interval to obtain the instability evolution acceleration, so as to predict the trend intensity of the evolution of the instability trajectory. The evolution direction is determined by the sign of the difference in the state instability characteristic values. If the difference is positive, it is determined to be a divergent direction; if the difference is negative, it is determined to be a convergent direction. When the evolution direction is divergent, a dynamic weighting coefficient is calculated based on the magnitude of the instability evolution acceleration. This dynamic weighting coefficient is the sum of the preset benchmark weight value and the absolute value of the instability evolution acceleration. The product of the dynamic weighting coefficient and the instability evolution velocity is then calculated, and the result of this product is used as the driving risk urgency index. When the evolution direction is convergent, the driving risk urgency index is set to zero or a preset safety value. The driving risk urgency index is matched with a preset multi-level risk threshold range, and the vehicle driving risk level is output according to the numerical range of the matching result.

7. The method for predicting vehicle driving risk status according to claim 6, characterized in that, S6 matches the corresponding boundary correction coefficient according to the determined vehicle driving risk level, and uses the boundary correction coefficient to dynamically correct the vehicle dynamic stability boundary conditions, so as to define the boundary of the safe and feasible domain in the multi-dimensional driving state space. Obtain the coordinate position of the vehicle's trajectory point in the multi-dimensional driving state space at the current moment; calculate the shortest Euclidean distance between the current trajectory point and the boundary of the safe and feasible domain; define the shortest Euclidean distance as the stability margin, which refers to the remaining safe distance of the vehicle driving dynamics system from the instability critical point.

8. The method for predicting vehicle driving risk status according to claim 7, characterized in that, S7 determines the target recovery point within the safe and feasible region along the shortest path based on the current trajectory point and the boundary of the safe and feasible region in the multi-dimensional driving state space; Calculate the multidimensional coordinate difference between the target recovery point and the current trajectory point, and generate a state correction vector. The state correction vector represents the direction and magnitude of the state parameter adjustment required for the vehicle to recover to the safe and feasible region. Integration The vehicle driving risk level, stability margin value, and state correction vector are used to construct a multi-dimensional driving risk state assessment result; The multi-dimensional driving risk status assessment results include risk level prompt information, stability margin values, and status correction vectors, wherein the risk level prompt information is used to drive the vehicle display terminal to provide visual warnings.

Images