A new energy vehicle safe driving control method and system

Abstract

The present application relates to the technical field of driving control, and particularly relates to a new energy vehicle safe driving control method and system. The method comprises the following steps: obtaining a historical emergency avoidance data set and a corresponding user misoperation braking state through a new energy vehicle control center, first performing sample undersampling processing to obtain an emergency avoidance balanced sample. Then, the balanced sample is used to analyze user misoperation braking abnormalities, and based on the braking abnormality data, an unordered vector intensity calculation in the time dimension is performed to generate unordered vector segmented clustering data. Subsequently, braking timing intervention correction perception is performed according to the clustering data to generate optimized braking intervention correction data. Finally, the optimized data is transmitted to a new energy vehicle control terminal to perform safe driving control. The present application optimizes the driving control technology to make the driving control technology more perfect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of driving control, in particular to a new energy vehicle safe driving control method and system. BACKGROUND

[0002] New energy vehicles have higher complexity in power response, braking system, electronic control system and intelligent driving assistance system, which makes the driving operation in emergency more susceptible to driver misoperation. For example, in the emergency escape scene, the driver overbrakes, underbrakes or makes wrong operation due to unexpected situations, which not only reduces the efficiency of escape, but also may cause secondary accidents. Therefore, developing a safe driving method that can quickly perceive driver misoperation, analyze braking abnormalities and optimize control through intelligent intervention in emergency has become an important direction of new energy vehicle safety technology research. This method not only needs to model the driver's behavior based on historical driving data and emergency escape scenes, but also needs to use data analysis and clustering algorithms to identify the braking disorder mode and dynamically evaluate the braking behavior in the time dimension, so as to generate accurate correction strategies. However, the traditional new energy vehicle safe driving control method has the problem of inaccurate analysis of the braking abnormality intensity caused by user misoperation, resulting in large braking intervention errors. SUMMARY

[0003] Therefore, it is necessary to provide a new energy vehicle safe driving control method and system to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a new energy vehicle safe driving control method, the method comprising the following steps:

[0005] Step S1: obtaining a historical emergency escape data set and a corresponding user misoperation braking state through a new energy vehicle control center; performing sample undersampling processing on the historical emergency escape data set to obtain an emergency escape balanced sample;

[0006] Step S2: analyzing the user misoperation braking abnormality of the misoperation braking state according to the emergency escape balanced sample in the emergency escape scene to obtain braking abnormality data; performing braking disorder vector intensity evolution in the time dimension based on the braking abnormality data to obtain disorder vector segmented clustering data;

[0007] Step S3: performing braking timing intervention correction perception according to the disorder vector segmented clustering data to generate braking intervention correction optimization data; sending the braking intervention correction optimization data to a new energy vehicle control terminal to perform new energy vehicle safe driving control.

[0008] Preferably, the application further provides a new energy vehicle safe driving control system for executing the new energy vehicle safe driving control method as described above, the new energy vehicle safe driving control system comprising:

[0009] A data sampling module is configured to acquire a historical emergency avoidance data set and a corresponding user misoperation braking state through a new energy vehicle control center, perform sample undersampling processing on the historical emergency avoidance data set, and obtain an emergency avoidance balanced sample.

[0010] A braking abnormality analysis module is configured to perform misoperation braking abnormality analysis of the user misoperation braking state according to the emergency avoidance balanced sample to obtain braking abnormality data, and perform braking disorder vector strength calculation in a time dimension based on the braking abnormality data to obtain disorder vector segmented clustering data.

[0011] An intervention correction perception module is configured to perform braking timing intervention correction perception according to the disorder vector segmented clustering data to generate braking intervention correction optimization data, and send the braking intervention correction optimization data to a new energy vehicle control terminal to execute new energy vehicle safe driving control.

[0012] The beneficial effects of this invention are that by acquiring historical emergency avoidance datasets and corresponding user-incorrect braking states through the new energy vehicle control center, it is possible to effectively collect data on erroneous actions that occur during actual vehicle operation, providing real and reliable samples for subsequent analysis. By performing undersampling on the dataset, the imbalance between positive and negative samples can be balanced, ensuring that model training is not affected by bias. The generated balanced emergency avoidance samples can better represent the actual performance of drivers in emergency situations, providing basic data for subsequent analysis of abnormal braking actions, thus laying a solid foundation for optimizing safe driving control strategies. Based on the balanced emergency avoidance samples, by analyzing the braking states and identifying abnormal braking data, driver errors and abnormal behaviors in emergency scenarios can be accurately captured. This process not only reveals the driver's reaction characteristics in the face of emergency situations but also provides detailed abnormal data support for subsequent calculations of disordered braking vector strength. Calculations based on this data over time can accurately assess the degree of disorder in braking behavior at different time points, further generating disordered vector segmentation clustering data, providing rich information support for subsequent optimized control, and ensuring that driver errors are effectively identified and handled. This invention utilizes unordered vector segmented clustering data for braking timing intervention and correction perception, enabling real-time optimization of vehicle braking control based on specific driver erroneous actions. By accurately sensing the temporal changes in braking behavior, this process automatically identifies critical moments requiring intervention and correction, adjusting the braking system's operation at these times to prevent potential accidents. The generated braking intervention and correction optimization data effectively guides the vehicle control terminal to execute safer and more precise driving controls, helping drivers adjust in a timely manner for a safer driving experience. The implementation of this technology effectively improves the active safety performance of new energy vehicles, reduces accident risks, and enhances overall driving safety. Therefore, this invention optimizes a traditional new energy vehicle safe driving control method, addressing the problem of inaccurate analysis of braking anomalies caused by user erroneous actions, leading to large braking intervention errors. It improves the accuracy of analyzing braking anomalies caused by user erroneous actions and reduces the errors caused by braking intervention. Attached Figure Description

[0013] Figure 1 A flowchart illustrating the steps of a safe driving control method for new energy vehicles;

[0014] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0015] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S3. DETAILED DESCRIPTION

[0016] Referring to Figures 1 to 3 The application discloses a new energy vehicle safe driving control method, and the method comprises the following steps:

[0017] Step S1: obtaining a historical emergency avoidance data set and a corresponding user misoperation braking state through a new energy vehicle control center; performing sample undersampling processing on the historical emergency avoidance data set to obtain an emergency avoidance balanced sample;

[0018] Step S2: performing user misoperation braking anomaly analysis on the misoperation braking state according to the emergency avoidance balanced sample to obtain braking anomaly data; and performing braking disorder vector strength calculation in the time dimension based on the braking anomaly data to obtain disorder vector segmented clustering data;

[0019] Step S3: performing braking timing intervention correction perception according to the disorder vector segmented clustering data to generate braking intervention correction optimization data; and sending the braking intervention correction optimization data to a new energy vehicle control terminal to perform new energy vehicle safe driving control.

[0020] In the embodiment of the application, reference Figure 1 The application discloses a new energy vehicle safe driving control method, and the method comprises the following steps:

[0021] Step S1: obtaining a historical emergency avoidance data set and a corresponding user misoperation braking state through a new energy vehicle control center; performing sample undersampling processing on the historical emergency avoidance data set to obtain an emergency avoidance balanced sample;

[0022] In the embodiment of the application, the new energy vehicle control center extracts the emergency escape event original log data set stored in the past 36 months from the vehicle data recording module, which contains the timestamp of each event, accurate to millisecond level, with a sampling frequency of 100Hz, and records the vehicle longitudinal acceleration, lateral acceleration, steering wheel angle, throttle pedal opening, brake pedal pressure, vehicle speed, GPS longitude and latitude, and vehicle body stability control system ESP activation state. At the same time, the user misoperation brake state label bound with each emergency escape event is extracted, which is labeled by the artificial review labeling system within 72 hours after the event occurs. The label is based on video monitoring playback and sensor data cross verification. The historical emergency escape data set is subjected to sample undersampling processing. First, the positive and negative class distribution ratio in all samples is counted. It is found that the misoperation brake state is abnormal, accounting for only 8.7% of the total sample size, which belongs to a serious unbalanced data set. Tomek Links algorithm is used to identify and remove boundary overlapping samples. Edited Nearest Neighbors ENN method is used to remove noise points and ambiguity points. Finally, the number of negative samples is 23% of the original negative samples, and the positive and negative sample ratio is 1 to 1.05, forming an emergency escape balanced sample. The total number of the sample is 4728, each sample contains a time series length of 5s, a total of 500 sampling points, and each sampling point contains a 12-dimensional feature vector, including acceleration vector components, direction angle change rate, pedal opening gradient and other physical parameters.

[0023] Step S2: According to the emergency escape balanced sample, the misoperation brake state is analyzed for user misoperation brake abnormality in the emergency escape scene to obtain brake abnormality data; based on the brake abnormality data, the brake disorder vector intensity calculation in the time dimension is performed to obtain the disorder vector segmentation clustering data;

[0024] In the embodiment of the application, based on the emergency escape balance sample, three-dimensional scene restoration data is constructed, vehicle dynamics parameters (including vehicle mass 1560 kg, mass center height 0.52 m, wheelbase 2.7 m, tire lateral stiffness 85000 N / rad) are used, combined with GPS trajectory coordinates and IMU attitude angle data to reconstruct the spatial topology structure of the escape scene corresponding to each sample, and the misoperation braking state is simultaneously mapped to the time slice position in the scene to generate misoperation mapping data matrix of the escape scene, the dimension of the matrix is 4728 rows by 500 columns, each column corresponds to the operation state code value in the 10 ms time window, abnormal analysis is performed on the mapping data, the accelerator acceleration state is extracted, if the longitudinal acceleration value is greater than 12 m / s² for 1.5 s, and the accelerator pedal opening degree jumps from 30% to more than 90% within 100 ms, it is marked as accelerator misoperation abnormal acceleration state, the steering operation state is extracted, if the steering wheel angle cumulative change is more than 90 degrees within 1000 ms, and the direction switching action angle difference is greater than 45 degrees within any 333 ms window for 3 times or more, it is marked as direction misoperation abnormal steering, the two types of abnormal states are combined to generate braking abnormal data, the data structure is a binary mask sequence with a length of 500 bits, each bit corresponds to an abnormal trigger flag bit in a 10 ms time window, then the braking abnormal data is subjected to out-of-order vector intensity calculation in the time dimension, first, the acceleration change gradient in the abnormal time period is extracted, the acceleration increment in each 200 ms window is calculated, and the difference value of the previous window is formed to form a speed increase proportion difference sequence, the maximum swing amplitude of the steering wheel angle in the same window is calculated to form a steering angle mutation swing sequence, the speed increase proportion difference sequence and the steering angle mutation swing sequence are input into a yaw impact vector intensity analyzer, nonlinear regression fitting is performed, and after smoothing the data by using cubic spline interpolation, a yaw impact vector regression intensity sequence is output, the sampling rate of the sequence is 5 Hz, and there are 25 data points, each data point includes a two-dimensional vector composed of a longitudinal impact component and a lateral swing component, K-means clustering algorithm is performed on the regression intensity sequence, the clustering number K is set to 5, and the clustering characteristics are three dimensions of time position, intensity amplitude change slope, and a segmented clustering data of unordered vectors is generated, and 5 cluster clusters are output, each cluster includes a time start and end point center intensity vector and a cluster sample standard deviation.

[0025] Step S3: performing braking timing intervention correction perception according to the segmented clustering data of unordered vectors to generate braking intervention correction optimization data; and sending the braking intervention correction optimization data to a new energy vehicle control terminal to perform new energy vehicle safe driving control.

[0026] In the embodiment of the application, after receiving the unordered vector segment clustering data, firstly, one-dimensional convolution operation is performed on the time sequence segment of each cluster, the convolution kernel length is 5, the step length is 1, the weight is initialized as Gaussian distribution with mean 0 and variance 0.1, the bias term is fixed as 0, and the unordered vector segment convolution data is output after ReLU activation function, the data dimension is 25 rows multiplied by 3 columns, which respectively represent time index, vertical intensity after convolution and horizontal intensity after convolution, the segment direction offset angle is extracted from the convolution data, the angle is obtained by converting the horizontal intensity value through the inverse tangent function, and the unit is radian, the speed loss value is synchronously extracted, the value is obtained by subtracting the original speed signal from the vertical intensity value, and the unit is m / s, the body offset angle is quantized for each group of direction offset angle and speed loss value, and the table lookup method is used to map to the preset offset level table, the table includes 9 levels from negative 4 to positive 4, each level corresponds to an interval of 0.15 radian, the braking torque demand mapping table is queried according to the quantization result, the table pre-stores ideal braking torque values corresponding to different offset levels, and the unit is N·m, braking torque demand data is generated, smoothing control optimization is performed on the braking torque demand data, firstly, the direction offset intervention correction angle is obtained according to the current time period, the angle is equal to the original offset angle multiplied by the damping coefficient 0.7 to obtain the corrected angle value, the intervention correction speed is synchronously calculated, the speed is equal to the original speed minus the vertical intensity value multiplied by the compensation factor 0.3, and the unit is m / s, vehicle intervention correction braking jitter offset simulation calculation is performed by using the corrected angle and the corrected speed, four-order Runge-Kutta method is used to integrate the vehicle two-degree-of-freedom dynamics equation, the time step is set as 0.01s, the integration interval is the length of the current time period, and jitter offset data is output, the data includes body yaw rate and lateral acceleration, multivariate linear regression is performed on the jitter offset data, the independent variables are the corrected angle and the corrected speed, the dependent variables are the yaw rate and the lateral acceleration, the regression coefficient matrix dimension is 2 rows multiplied by 2 columns, the jitter offset regression data is obtained by solving through the least square method, progressive control is performed on the corrected angle and the corrected speed based on the regression data, the control law is the current value, which is equal to the value at the last moment plus the step length 0.02 radian, or 0.5 m / s multiplied by the sign function, the offset progressive correction angle and the progressive correction speed are output, and the two are substituted into the braking torque smoothing controller, the controller adopts the PID structure, the proportional gain is set as 12, the integral time constant is 0.8s, and the differential time constant is 0.05s, and the braking force smoothing control data is output, the data is a time sequence with a sampling rate of 10Hz and a length of 50 points, each point is a scalar torque value with a unit of N·m, and the braking force smoothing control data is input into the iterative learning module as the braking intervention correction data, the learning rate is set as 0.1, the iteration number is fixed as 5 rounds, and the gradient descent method is used for each round update, and the loss function is the mean square error threshold.01N·m², the brake intervention correction optimization data is generated, the data is in JSON structure, a time stamp array and a torque instruction array are included, and the data is sent to a new energy vehicle control terminal through a CAN bus protocol at a baud rate of 500 kbps, and after the terminal receives the data, the data is immediately written into a brake actuator control register address 0x3A2B, a hardware interrupt is triggered to start a hydraulic regulating valve to execute corresponding brake force distribution to realize safe driving control.

[0027] Step S1 includes the following steps:

[0028] Step S11: obtaining a historical emergency avoidance data set and a corresponding user misoperation brake state through a new energy vehicle control center;

[0029] Step S12: randomly sampling the historical emergency avoidance data set to obtain an emergency avoidance random sample;

[0030] Step S13: embedding a time label into the emergency avoidance random sample to obtain an avoidance time label sample;

[0031] Step S14: performing sample undersampling processing on the avoidance time label sample to obtain an emergency avoidance balanced sample.

[0032] In the embodiment of the application, the new energy vehicle control center accesses the vehicle-mounted data storage unit, the unit adopts eMMC 5.1 standard, the capacity is 32 GB, the data partition format is FAT32, the file system saves the driving event log in the last 36 months in a cyclic covering manner, the emergency avoidance data set is derived from the ESP electronic stability program trigger event log, each record contains a time stamp accurate to the millisecond level, the sampling frequency is fixed at 100 Hz, and the record field includes longitudinal acceleration, lateral acceleration, yaw rate, steering wheel angle, accelerator pedal position percentage, brake pedal pressure value unit kPa, vehicle speed unit m / s, GPS longitude and latitude coordinate accuracy 0.00001 degrees, and the corresponding user misoperation brake state is extracted from the driver behavior analysis module, the state is determined by an artificial annotation process completed within 72 hours after the event, the annotation is based on frame-by-frame interpretation of the synchronous alignment of the in-vehicle DVR video stream and the IMU sensor data, the annotation result is stored in binary encoding form, 1 represents misoperation, and 0 represents normal operation, the annotation process complies with the SAE J2945 / 1 standard, the annotators hold ISO 39001 certification, the annotation consistency is verified by a Kappa coefficient greater than 0.89, and the original data set includes 12456 records, each record is 5s long, a total of 500 sampling points, and each sampling point contains 14 physical quantity fields.

[0033] A random sample selection operation is performed on the historical emergency avoidance dataset, a Mersenne Twister pseudo-random number generator is used, the seed is fixed as 0x7A3F, the random number sequence generator is initialized, the output range is limited between 1 and 12456, the integer index value is not repeated, the extraction quantity is set as 40% of the original dataset, that is, 4982 records, the extraction process is carried out in batches, 500 records are processed in each batch to avoid memory overflow, each selected record completely retains its 500 sampling point time sequence structure and all 14-dimensional feature fields, forms an emergency avoidance random sample, the sample maintains the time continuity and spatial coordinate integrity of the original data, and does not perform any interpolation or truncation operation, the sample distribution is subjected to Shapiro-Wilk normality test, and the p value is greater than 0.05, indicating that the randomness meets the uniform distribution assumption, the sample index list is exported in CSV format, the first column is the original record ID, the second column is the extraction timestamp, and the third column is the last six digits of the vehicle VIN code for traceability verification.

[0034] A time label embedding operation is performed on the emergency avoidance random sample, first, the timestamp field embedded in each sample is parsed, the format is YYYY-MM-DD HH:MM:SS.mmm, seven segment values of year, month, day, hour, minute, second and millisecond are extracted, and are respectively converted into integer variables to construct a seven-dimensional time feature vector, the time feature vector is respectively attached to the 500 sampling points of each sample to form an extended data structure, the new structure dimension is 500 rows multiplied by 21 columns, and seven new columns are added as time label fields, including year field value range 2020 to 2024, month field value range 1 to 12, date field value range 1 to 31, hour field value range 0 to 23, minute field value range 0 to 59, second field value range 0 to 59, and millisecond field value range 0 to 999, in order to ensure time continuity, a sliding window cumulative correction is performed on the millisecond field, the rule is that if the current sampling point millisecond value is less than the previous point, 1000 is automatically added and the minute field is borrowed, when the second field overflows, it is advanced to the minute field, and so on, finally, an avoidance time label sample is output, the total number of entries is still 4982, the data block size of each sample is 21000 bytes, and the storage format is HDF5.

[0035] The sample undersampling processing is performed on the emergency-avoiding time label sample. Firstly, the user misoperation braking state label distribution is counted. The result shows that the label 1, i.e. the abnormal operation sample, is 433, the label 0, i.e. the normal operation sample, is 4549, the class ratio is 1 to 10.5, the Tomek Links algorithm is used to identify the boundary overlapping sample pairs, the Euclidean distance between all sample pairs is calculated, the threshold is set to 0.35 standard deviation units, 217 sample pairs with different labels and the nearest neighbors are identified, and all are removed. Then, the Edited Nearest Neighbors (ENN) algorithm is applied, the neighbor number k is set to 3, and the 3 nearest neighbors of the remaining samples are checked point by point to see if the majority class is consistent with itself. If not, it is removed. A total of 89 noise points are removed, finally 433 label 1 samples and 454 label 0 samples are retained, the positive and negative sample ratio reaches 1 to 1.048, the emergency-avoiding balanced sample is formed, the total amount of the sample is 887, each sample maintains a 21-dimensional feature structure, the time label is completely retained, and no normalization or standardization processing is performed.

[0036] Step S2 comprises the following steps:

[0037] Step S21: restoring the emergency-avoiding scene according to the emergency-avoiding balanced sample; and simultaneously, performing associated mapping on the misoperation braking state according to the emergency-avoiding scene to obtain emergency-avoiding scene-misoperation mapping data;

[0038] Step S22: performing user misoperation braking abnormality analysis on the emergency-avoiding scene of the emergency-avoiding scene-misoperation mapping data to obtain braking abnormality data;

[0039] Step S23: performing braking disorder vector intensity calculation in the time dimension based on the braking abnormality data to obtain braking disorder vector intensity;

[0040] Step S24: performing time sequence segmentation clustering on the braking disorder vector intensity to obtain disorder vector segmentation clustering data.

[0041] As an example of the present application, reference is made to Fig. 1 showing that in the present example, the step S2 comprises: Figure 2

[0042] Step S21: restoring the emergency-avoiding scene according to the emergency-avoiding balanced sample; and simultaneously, performing associated mapping on the misoperation braking state according to the emergency-avoiding scene to obtain emergency-avoiding scene-misoperation mapping data;

[0043] ​In the embodiment of the application, the time series data extracted from each record in the emergency escape balance sample includes eight core physical quantity fields of longitudinal acceleration, lateral acceleration, yaw rate, steering wheel angle, throttle pedal opening, brake pedal pressure, vehicle speed, and GPS coordinates, combined with the inherent parameters of the vehicle, such as the total vehicle mass of 1560 kg, the distance from the mass center to the front axle of 1.3 m, the distance from the mass center to the rear axle of 1.4 m, the tire cornering stiffness of the front wheel of 85000 N / rad, the tire cornering stiffness of the rear wheel of 78000 N / rad, the air resistance coefficient of 0.29, the windward area of 2.1 square meters, and the rolling resistance coefficient of 0.015, to construct a three-dimensional kinematics scene restoration data, to use the rigid body six-degree-of-freedom motion equation to express the attitude change in Euler angles, to fix the time step at 0.01 s, to use the fourth-order Runge-Kutta method for integration, to set the initial condition by the sensor readings at the starting time of the sample, and to output the vehicle body position XYZ coordinates updated every 10 ms, with the unit of m, the vehicle body yaw angle with the unit of radian, and the vehicle body pitch angle with the unit of radian. The misoperation brake state label is extracted from the sample synchronously, the label is a binary value of 1 or 0, the label is aligned with time and embedded into the 9th column of the scene matrix to form the escape scene misoperation mapping data, the data structure is 500 rows by 9 columns, each row corresponds to a 10 ms time window, the first to eighth columns are physical state quantities, and the ninth column is the operation state label. A total of 887 samples are all subjected to the same processing procedure without omission, interpolation, or extrapolation, all numerical values are kept at the original sensor accuracy, the longitudinal acceleration is kept to three decimal places with the unit of m / s², the steering wheel angle is kept to an integer with the unit of degrees, and the vehicle speed is kept to two decimal places with the unit of m / s.

[0044] Step S22: performing user misoperation brake abnormality analysis on the escape scene-misoperation mapping data to obtain brake abnormality data;

[0045] In the embodiment of the application, the user misoperation brake abnormality analysis of the risk avoidance scene misoperation mapping data is performed as follows: first, 500 rows of data of each sample are scanned to extract the throttle acceleration state, the determination condition is that the longitudinal acceleration value is greater than 12 m / s2 in 150 consecutive rows, i.e. 1.5 s, and the throttle pedal opening degree jumps from less than 30% to greater than 90% in any 10 consecutive rows, i.e. 100 ms, and the above two conditions are met at the same time, then the current time window is marked as throttle misoperation abnormality, the acceleration state marker value is 1, otherwise it is 0; second, the steering operation state is extracted, the determination condition is that the absolute value of the cumulative change amount of the steering wheel angle is greater than or equal to 90 degrees in any 100 row window, i.e. 1 s window, and there are not less than 3 times in the window, and the absolute value of the angle difference of each switching angle is greater than 45 degrees, and the direction misoperation abnormality is marked if the above conditions are met, the steering marker value is 1, otherwise it is 0; the throttle misoperation abnormality and acceleration state marker sequence and the direction misoperation abnormality and steering marker sequence are logically ORed to generate the final brake abnormality data, which is a binary sequence with a length of 500, each bit corresponds to an abnormality trigger state of a 10 ms time window, 1 represents abnormality, and 0 represents normality.

[0046] Step S23: performing brake disorder vector strength calculation in time dimension based on the brake abnormality data to obtain brake disorder vector strength;

[0047] In the embodiment of the application, the brake disorder vector strength calculation in time dimension based on the brake abnormality data is performed as follows: first, all time window indexes marked as 1 are located to form an abnormal time period list, the longitudinal acceleration sequence of each abnormal time period is extracted, the difference between each adjacent two points is calculated to obtain the acceleration gradient sequence, and the difference between the maximum value and the minimum value of the gradient sequence is calculated to obtain the acceleration gradient difference unit m / s3, the steering wheel angle sequence in the same time period is extracted synchronously, the standard deviation of the steering wheel angle sequence is calculated to obtain the steering angle mutation swing unit degree, and the acceleration gradient difference and the steering angle mutation swing are taken as inputs to perform yaw impact vector strength analysis, the two groups of data are respectively interpolated by using cubic spline interpolation, the interpolation node interval is set as 0.2 s, the number of data points after interpolation is uniformly set as 25, and the non-linear regression fitting function is selected to be a Legendre polynomial with an order of 3, the yaw impact vector regression strength sequence is output after fitting, the sequence includes 25 two-dimensional vectors, the first component of each vector is the longitudinal impact strength, the unit is N, and the second component is the lateral swing strength, the unit is N·m, the sliding window standard deviation calculation is performed on the regression strength sequence, the window length is 5 points, the step length is 1 point, and the brake disorder vector strength is output, the length is 21 points, and each point is a scalar value with a unit of N·m·s, which represents the local dynamic disorder degree.

[0048] Step S24: performing time sequence segmentation clustering on the brake disorder vector strength to obtain disorder vector segmentation clustering data.

[0049] In the embodiment of the application, the time sequence segmentation clustering is performed on the braking disorder vector intensity sequence. First, a clustering feature vector corresponding to each time point is constructed, which contains three items. The first item is a time index value, ranging from 1 to 21. The second item is a current point intensity value. The third item is an intensity difference value with the previous point. If it is the first point, the difference value is 0. A feature matrix of 21 rows by 3 columns is formed. The K-means clustering algorithm is performed on the matrix. The number of clusters K is fixed at 5. The initialization method is Forgy, which randomly selects 5 sample points as initial center points. The Manhattan distance is used for distance measurement. The iteration termination condition is that the cluster center movement distance is less than 0.001 for three consecutive rounds or the maximum iteration number reaches 100 rounds. After clustering is completed, 5 clusters are output. Each cluster contains a list of time points belonging to the cluster, a cluster center coordinate triple, and an average Manhattan distance value of the samples in the cluster to the center. The time points of each cluster are arranged in ascending order, and adjacent time points are merged to form continuous time periods. The starting index and the ending index are marked. The disorder vector segmentation clustering data is generated. The data structure is 5 rows. Each row contains 5 fields, including the segmented starting time point number, the segmented ending time point number, the segmented center intensity value, the segmented duration length (unit: s), and the segmented intensity fluctuation standard deviation. The clustering process is independently performed on all 887 samples.

[0050] Step S22 includes the following steps:

[0051] Extracting the accelerator acceleration state and the direction steering operation state of the user operation in the risk avoidance scene-misoperation mapping data;

[0052] When the acceleration of the accelerator acceleration state is greater than 12 , the duration is greater than 1.5 s, and the accelerator pedal opening degree change is extracted through the engine control unit, when the accelerator pedal opening degree change is greater than 60% in 1 s from the increasing rate, it is determined as an accelerator misoperation abnormal acceleration state;

[0053] When the direction steering operation state is greater than or equal to 90°~180° in 1 s, and the direction steering operation frequency is greater than or equal to 3 times of the direction steering operation state in 1 s, it is determined as a direction misoperation abnormal steering;

[0054] Based on the accelerator misoperation abnormal acceleration state and the direction misoperation abnormal steering, the user misoperation brake abnormality analysis of the risk avoidance scene is performed to obtain the brake abnormality data.

[0055] In the embodiment of the application, the accelerator pedal state and the steering state of the user operation in the risk avoidance scene-misoperation mapping data are realized by using a multi-sensor fusion technology. First, original data is extracted from a vehicle control area bus network. A CAN-FD communication protocol is used, the baud rate is 2 Mbps, the frame format is an extended frame, the identifier length is 29 bits, the maximum length of the data field is 64 bytes, and the data sampling frequency is set to 200 Hz to ensure the capture of transient operation characteristics. Wheel speed sensor data is obtained from a vehicle chassis domain controller, with an accuracy of 0.01 km / h. Acceleration sensor and gyroscope data is obtained from a vehicle body stability control system, with accuracies of 0.005 g and 0.01 degrees / second, respectively. Steering wheel angle and torque sensor data is obtained from a steering system controller, with accuracies of 0.1 degrees and 0.02 N·m, respectively. Throttle pedal position sensor data is obtained from a powertrain controller, with an accuracy of 0.1%. A signal denoising algorithm is applied to the extracted original data. A wavelet denoising technology is used. Daubechies wavelets are used. The number of decomposition layers is 4. The threshold value selection uses Stein unbiased risk estimation. The threshold value reduction method is a soft threshold value. The signal-to-noise ratio is improved by more than 15 dB. The denoised data is subjected to feature extraction. An accelerator pedal state feature vector is calculated, including five parameters: throttle pedal position, throttle pedal speed, throttle pedal acceleration, vehicle longitudinal acceleration, and engine speed. A sliding window method is used to calculate the statistical characteristics of these parameters at different time scales. The window sizes are 50 ms, 100 ms, 200 ms, 500 ms, and 1000 ms. The features include mean, standard deviation, peak value, valley value, and root mean square value. The same method is used to extract a steering state feature vector from the steering state, including five parameters: steering wheel angle, steering wheel angular velocity, steering wheel angular acceleration, vehicle yaw rate, and lateral acceleration. Statistical characteristics are calculated according to the same time window. Principal component analysis is applied to the extracted feature vectors for dimension reduction. Principal components that explain 95% of the variance are retained. The dimension of the reduced feature vector is reduced from 125 to 32. Data compression is realized while retaining key information. Finally, a feature data set of the accelerator pedal state and the steering state is formed, including 22,051 records. Each record includes a 32-dimensional feature vector and timestamp information, providing a data basis for subsequent abnormal state determination.

[0056] When the acceleration of the accelerator acceleration state is greater than 12 m / s², the duration is greater than 1.5 s, and the accelerator pedal opening degree change is extracted through the engine control unit at the same time, when the accelerator pedal opening degree change is greater than 60% in 1 s, it is determined that the accelerator misstep abnormal acceleration state is determined, the determination process is realized by a multi-threshold joint decision mechanism, first, the vehicle longitudinal acceleration is measured by a high-precision acceleration sensor, the range is ±16g, the resolution is 0.002g, the sampling frequency is 200Hz, the acceleration signal is processed by a second-order Butterworth low-pass filter, the cutoff frequency is 25Hz, and high-frequency noise is filtered out, an acceleration threshold judgment module is established, and the acceleration threshold is set to , which is determined by statistical analysis of 12580 emergency avoidance events, which is outside the boundary of normal acceleration distribution, the system continuously monitors the filtered acceleration signal, starts the timer when the acceleration exceeds the threshold, the timing accuracy is 1ms, continuously monitors the acceleration value, until the acceleration drops below the threshold or the timer reaches the set time, the set duration threshold is 1.5s, which is determined based on human reaction time and vehicle dynamics characteristics, at the same time, the system reads the accelerator pedal opening degree data through the OBD-II interface of the engine control unit, the interface protocol conforms to the ISO15765-4 standard, the baud rate is 500kbps, the reading frequency is 50Hz, the accelerator pedal opening degree sensor is a Hall effect type, the accuracy is 0.2%, the range is 0-100%, the system records the accelerator pedal opening degree change in 1s time window, calculates the opening degree change rate by discrete difference, the difference step is 20ms, when the maximum change rate exceeds 60% / s, the accelerator incremental rate abnormal flag is triggered, the threshold is determined by comparing the accelerator operation characteristics in normal driving and emergency avoidance events, 95% of the accelerator operation change rate in normal driving is less than 40% / s, the system uses and gate logic to judge three conditions: acceleration greater than 12 m / s², duration greater than 1.5s, accelerator pedal opening degree change rate greater than 60% / s, when the three conditions are met at the same time, the system determines that the accelerator misstep abnormal acceleration state is determined, the determination result is stored in the vehicle event recorder in the form of a Boolean value, at the same time, the time stamp, vehicle speed, acceleration and accelerator position and other key parameters when triggered are recorded, forming an abnormal event record, providing a basis for subsequent brake abnormality analysis.

[0057] When the direction turning operation state is greater than or equal to 90°~180° within 1s, and the direction turning operation frequency is greater than or equal to 3 times within 1s, the direction misoperation abnormal turning is determined. The determination process is realized by using time-frequency analysis technology. First, the steering angle data is collected by the steering wheel angle sensor of the electric power steering system. The sensor type is an optical encoder, the resolution is 0.1 degrees, the range is ±720 degrees, and the sampling frequency is 100 Hz. The original angle signal is filtered by a median filter to remove burrs, and the window size is 5 sampling points. The filtered signal is smoothed by a first-order low-pass filter, and the cutoff frequency is 15 Hz. The system establishes a steering wheel angle change detection module, and the sliding window method is used to calculate the steering wheel angle change range within 1s. The window size is 100 sampling points, and the step size is 1 sampling point. The difference between the maximum angle value and the minimum angle value is calculated for each window to obtain the angle change amount. The angle change threshold is set to [90°, 180°], and the threshold interval is determined by statistical analysis of 6450 cases of direction misoperation. When the angle change amount in the window falls within the threshold interval, the angle change abnormal flag is triggered. At the same time, the system uses fast Fourier transform to analyze the frequency characteristics of the steering wheel angle signal. The data in the 1s window is subjected to FFT, the window function is selected as Hanning window, and the FFT point number is set to 128. The main frequency component is extracted from the analyzed frequency spectrum, and the proportion of energy concentrated in the frequency band above 2 Hz is calculated. The direction turning operation frequency is determined by counting the zero-crossing points of the steering wheel angle signal. The threshold is set to 6 zero-crossing points within 1s, corresponding to an operation frequency of 3 times / s. The threshold is determined based on normal driving data analysis. In normal driving, 99% of the operation frequency is less than 2 times / s. The system uses a fuzzy logic controller to comprehensively evaluate the angle change and operation frequency parameters. The fuzzy controller input variables are the normalized angle change amount and operation frequency, and the output variable is the abnormality degree score, with a value range of 0-100. The fuzzy rule base contains 9 rules describing the relationship between angle change and operation frequency and abnormality degree. Mamdani reasoning method and barycenter method are used to solve the ambiguity. When the abnormality degree score exceeds 75 points, the system determines that the direction misoperation abnormal turning occurs. The determination result is recorded in the form of an event, including trigger time, vehicle speed, yaw angular velocity, steering wheel angle sequence and other key information, providing basic data for subsequent brake abnormality analysis.

[0058] The user misoperation brake abnormality analysis of the risk avoidance scene is performed based on the accelerator misstep abnormal acceleration state and the direction misoperation abnormal steering. A multi-feature fusion analysis method is adopted. First, a user misoperation brake abnormality feature library is constructed, including two main abnormal modes: brake abnormality based on accelerator misstep and brake abnormality based on direction misoperation. The feature library is established based on 8745 historical abnormal data. The hierarchical clustering algorithm is used to divide the abnormal data into 12 typical subclasses. The Ward minimum variance method is used for clustering, and the Mahalanobis distance is used for distance measurement. Each subclass is represented by a feature vector. The vector dimension is 24, including 16 time domain features and 8 frequency domain features. For a newly detected abnormal event, the same dimension feature vector is extracted. The minimum distance classifier is used to determine the category. The K nearest neighbor algorithm is used for the classifier, and the K value is set to 5. The cosine similarity is used for distance measurement. For the event judged as the accelerator misstep abnormal acceleration state, the system extracts the brake system response features, including brake pedal position, brake pressure, brake force distribution, and deceleration parameters. The relationship between brake intervention time and accelerator release time is analyzed. The brake delay time is calculated. The normal value range is 0.2-0.5 seconds. The abnormal value is greater than 0.8 seconds or less than 0.1 second. The time-frequency characteristics of the brake pedal displacement signal are analyzed by discrete wavelet transform. The Haar wavelet is used. The decomposition level is 3. The high-frequency coefficients are extracted to represent the stability of the pedal operation. For the event judged as the direction misoperation abnormal steering, the system analyzes the coordination between the steering wheel operation and the brake operation. The time difference between the steering wheel angular velocity peak value and the brake pressure peak value is calculated. The normal coordination operation time difference is 0.3-0.7 seconds. The abnormal value is greater than 1 second or less than 0.2 second. The matching degree of brake force distribution and steering demand is analyzed. The brake pressure difference between the left and right wheels is compared with the ideal brake pressure difference. The ideal value is calculated by the vehicle dynamics model. Based on the analysis results of the two types of abnormal states, the system establishes a brake abnormality classifier. The random forest algorithm is used. The number of decision trees is set to 100. The maximum depth is set to 8. The minimum leaf node sample number is 10. The feature random selection ratio is 0.7. The brake abnormality is divided into 6 categories: slight delay brake, severe delay brake, insufficient brake, excessive brake, uneven brake force distribution, and brake-steering interference. The classification accuracy is evaluated by 10-fold cross-validation. The accuracy rate reaches 92.3%. Finally, the brake abnormality data is generated, including the abnormal type label, the abnormal severity score, and the brake system key parameter time series data. The abnormal data is stored in a structured format, including header information and data segment. The header information records the event ID, timestamp, vehicle information, and abnormal type. The data segment records the complete process data from 3 seconds before the abnormal event to 2 seconds after the abnormal event at a sampling rate of 100 Hz, providing a data basis for the subsequent brake vector intensity calculation.

[0059] Step S23 includes the following steps:

[0060] Step S231: extract the accelerator misstep abnormal acceleration change and the direction misoperation abnormal steering change angle of the brake abnormal data in the time dimension;

[0061] Step S232: calculate the acceleration ratio of the accelerator misstep abnormal acceleration change, and then calculate the acceleration ratio difference in 2s time;

[0062] Step S233: analyze the steering angle mutation swing of the direction misoperation abnormal steering change angle in 2s time;

[0063] Step S234: analyze the yaw impact vector strength according to the acceleration ratio difference and the steering angle mutation swing, and obtain the yaw impact vector strength;

[0064] Step S235: perform nonlinear regression analysis on the yaw impact vector strength to obtain the yaw impact vector regression strength; and perform brake disorder vector strength calculation in the time dimension based on the yaw impact vector regression strength to obtain the brake disorder vector strength.

[0065] In the embodiment of the application, the accelerator misstep abnormal acceleration change and the direction misoperation abnormal steering change angle of the brake abnormal data in the time dimension are extracted, and a multi-source data synchronous analysis technology is adopted to realize the extraction. First, the system extracts the brake abnormal data from the vehicle control domain network CAN-FD bus. The data frame format conforms to the ISO 11898-1 standard, the baud rate is 5 Mbps, and the data contains the vehicle identification code, the time stamp and the abnormal identifier. The time synchronization protocol is used to ensure that the synchronization accuracy of the multi-sensor data reaches 1 ms. Then, the system extracts the accelerator misstep abnormal acceleration change record from the brake abnormal data. The record is collected by an acceleration sensor array. The sensor model is Bosch MM5.10, the range is ±16 g, the resolution is 0.0005 g, and the sampling rate is 500 Hz. The sensor array includes four measurement points in front, back, left and right, forming a redundant measurement structure. The vehicle centroid acceleration is calculated by a data fusion algorithm. The weighted average method is adopted in the fusion algorithm, and the weight coefficients are determined by offline optimization through the least square method. The optimization target is to minimize the fusion error, and the weight parameters are set to [0.35, 0.25, 0.2, 0.2]. The fused acceleration data is smoothed by a Kalman filter. The filter parameters are set to process noise covariance 0.01 and measurement noise covariance 0.08. The system extracts the direction misoperation abnormal steering change angle data. The data is collected by a steering wheel angle sensor. The sensor type is magneto-electric, the accuracy is 0.05°, the range is ±900°, and the sampling rate is 200 Hz. The angle signal is processed by a band-pass filter. The filter is a Butterworth third-order filter, the passband is 0.2 Hz-10 Hz, and the low-frequency drift and high-frequency noise are filtered out. The system organizes the extracted accelerator misstep abnormal acceleration data and direction misoperation abnormal steering angle data into a time series data structure. The fixed time window segmentation technology is adopted. The window length is set to 2.5 seconds, and the window overlap rate is 50%. Each window contains 1250 sampling points of accelerator misstep abnormal acceleration data and 500 sampling points of direction misoperation abnormal steering angle data. Feature extraction is performed on each time window. The statistical features include mean, standard deviation, maximum, minimum, peak-to-valley value, zero-crossing point number and autocorrelation coefficient. Each window generates a 32-dimensional feature vector. After dimension reduction by principal component analysis, 16 principal components are retained. The explained variance of the retained principal components is more than 98%. The system uses the dynamic time warping algorithm to align the time series of different samples, eliminates the influence of time scale change, and finally generates the aligned accelerator misstep abnormal acceleration change and direction misoperation abnormal steering change angle data sequence. The data amount is 8745 records, each record length is 2 seconds, the sampling rate is unified to 200 Hz, that is, 400 sampling points, which lays a foundation for subsequent speed increase ratio and steering swing amplitude analysis.

[0066] The rate of increase of abnormal acceleration changes due to throttle misapplication was calculated using a piecewise linear fitting method. First, the extracted abnormal acceleration data from throttle misapplication was divided into time windows, with each window size set to 50ms and adjacent windows overlapping by 10ms, resulting in 42 analysis windows over a 2-second time series. For each window, the linear trend of the acceleration data was calculated, and a straight line was fitted using the least squares method to obtain the slope value, representing the rate of change of acceleration within that window, in m / s³. The coefficient of determination (R²) of the fitted straight line must be greater than 0.85; otherwise, a quadratic curve fitting was used. The 42 acceleration data were then analyzed. The rate of change of acceleration was normalized, with a reference value set at 1.5 m / s³, a typical rate of change of acceleration under normal emergency acceleration conditions. The normalization formula used the ratio method, dividing the actual rate of change of acceleration by the reference value. The normalized value was defined as the growth rate ratio, which theoretically ranges from -∞ to +∞, but is actually concentrated in the range of -10 to +15. A time series of the growth rate ratio was established, containing 42 data points. A sliding window analysis was performed on this series, with a window size of 2 seconds to cover all 42 data points. The maximum and minimum values ​​of the growth rate ratio were calculated, and the difference between them was defined as... The growth rate ratio difference describes the fluctuation of acceleration change over a 2-second period, with a typical value range of 0-25. The system performs segmented analysis on the growth rate ratio difference, dividing the 2-second time into three stages: before, during, and after. Each stage contains 14 data points, and the mean growth rate ratio of each stage is calculated, denoted as [R_before, R_during, R_after]. The rate of change between the three stages is calculated and defined as [ΔR_before-during, ΔR_during-after], describing the time evolution characteristics of the growth rate ratio. The system establishes a growth rate characteristic vector, including the growth rate ratio difference, the three stage mean values, and the two rates of change, for a total of 6 dimensions. The system employs a weighted fusion method to comprehensively evaluate acceleration characteristics. The weight coefficients are determined through historical data analysis and set to [0.35, 0.15, 0.15, 0.15, 0.1, 0.1]. The weighted sum is calculated to obtain the comprehensive acceleration score, which ranges from 0 to 100. A higher score indicates a more dramatic change in acceleration. The system uses the acceleration ratio difference and the comprehensive acceleration score as key features and passes them to the yaw impact vector intensity analysis module to provide input parameters for subsequent yaw impact analysis. The calculation accuracy of the acceleration ratio difference reaches ±0.5, and the calculation accuracy of the comprehensive acceleration score reaches ±2.5.

[0067] The steering angle mutation swing of the abnormal steering change in the analysis direction is analyzed within 2s, which is realized by using a multi-scale spectrum analysis method. First, the steering wheel angle signal collected is preprocessed, and the median filter is used to remove abnormal spikes, with a window size of 5 sampling points. Then, the signal is processed by using a Savitzky-Golay smoothing filter, with a polynomial order of 3 and a window length of 9 sampling points, so as to remove high-frequency noise while retaining the turning characteristics of the signal. The system extracts 2s window data from the preprocessed steering wheel angle signal, with a total of 400 sampling points. The short-time Fourier transform is used to analyze the angle signal in time and frequency, with a Hamming window function, a window length of 64 points, and an overlap rate of 50%. Twelve time-frequency analysis segments are formed, with a frequency resolution of 3.125Hz. The frequency spectrum energy distribution characteristics of each time-frequency segment are extracted, and the 0.5Hz-5Hz frequency band is focused on. This frequency band contains the characteristic frequency of the driver's active steering operation. The system calculates the maximum change rate of the steering wheel angle, with a unit of degrees per second. The central difference method is used to calculate the angular velocity, with a difference step of 5ms. The maximum change rate usually exceeds 300 degrees per second in abnormal steering operation. The steering angle mutation swing is defined as the peak-to-peak value of the steering wheel angle within the 2s window, i.e. the maximum value minus the minimum value. The calculation method is to search for the global maximum and minimum values within 2s, and take the absolute difference value. The typical range of the mutation swing is 90 degrees-540 degrees. For the swing exceeding 360 degrees, the system analyzes the number of complete cycles of the steering wheel turning. The zero-crossing point counting method is used to identify the complete cycle, and the cycle number and the duration of each cycle are recorded. The system further analyzes the frequency characteristics of the steering operation, and uses the self-correlation analysis method with an adaptive window size between 100ms and 500ms to calculate the dominant frequency of the angle signal. The frequency range is 0.5Hz-10Hz. The normal steering operation frequency is usually lower than 1.5Hz, and the abnormal operation frequency can reach 3Hz-8Hz. The system calculates the differential entropy of the steering angle signal to quantify the uncertainty of the signal. The higher the entropy value, the more disordered the steering operation. The entropy value range of the normal steering operation is 1.2-2.5, and the entropy value range of the abnormal operation is 2.6-4.8. The system integrates the steering angle mutation swing, the maximum change rate, the operation frequency and the differential entropy of the four parameters to establish a steering abnormal feature vector. The feature vector is fused by using the weighted sum method, with a weight coefficient of [0.4, 0.25, 0.2, 0.15]. The steering abnormal comprehensive score is obtained, with a score range of 0-100.

[0068] The yaw impact vector strength analysis is realized by using a multi-physical quantity coupling analysis method according to the speed ratio difference and the steering angle mutation swing. First, a vehicle kinematics model is established, which contains 7 degrees of freedom, i.e. longitudinal motion, lateral motion, yaw motion and four wheel rotations. The model parameters are determined based on the physical characteristics of the vehicle, including vehicle mass 1850 kg, moment of inertia 2860 kg·m², wheelbase 2.78 m, front and rear wheel track 1.62 m and 1.63 m respectively, mass center height 0.56 m, the system maps the speed ratio difference to the longitudinal impact component, the mapping uses a piecewise linear function, when the speed ratio difference is less than 5, the longitudinal component = speed ratio difference x 4, when the speed ratio difference is between 5-15, the longitudinal component = 20 + speed ratio difference x 6, when the speed ratio difference is greater than 15, the longitudinal component = 80 + speed ratio difference x 2, the dimension of the longitudinal impact component is N·s, the theoretical range is 0-120, the system maps the steering angle mutation swing to the lateral impact component, the mapping also uses a piecewise linear function, when the mutation swing is less than 120 degrees, the lateral component = mutation swing x 0.25, when the mutation swing is between 120-300 degrees, the lateral component = 30 + (mutation swing-120) x 0.4, when the mutation swing is greater than 300 degrees, the lateral component = 102 + (mutation swing-300) x 0.1, the dimension of the lateral impact component is N·s, the theoretical range is 0-126, the system calculates the longitudinal-lateral coupling effect, the coupling coefficient is determined by the vehicle dynamics model, considering the influence of the mass center height, wheelbase, wheel track and road adhesion coefficient, the adhesion coefficient is divided into four typical values: dry asphalt road 0.85, wet asphalt road 0.6, snowy road 0.3 and ice surface 0.1, the corresponding coupling coefficients are 0.15, 0.25, 0.4 and 0.55 respectively, the coupling term is calculated as longitudinal component x lateral component x coupling coefficient, the system calculates the yaw impact vector strength, using the three-component synthesis method, the longitudinal component, lateral component and coupling term are weighted and summed, the weight coefficients are [0.35, 0.45, 0.2], the normalization coefficient is 0.01, the final yaw impact vector strength value range is 0-100, the system further analyzes the direction characteristics of the yaw impact vector, calculates the angle of the longitudinal and lateral components, the angle range is 0-90°, the larger the angle, the higher the proportion of the lateral component, the system decomposes the yaw impact vector strength into four sectors according to the direction: longitudinal dominant area (0-22.5°), longitudinal-lateral mixed area (22.5-45 degrees), lateral-longitudinal mixed area (45-67.5°) and lateral dominant area (67.5-90°), the impact characteristics of different sectors are obviously different, different braking control strategies need to be adopted, the system calculates the projection components of the yaw impact vector strength in the four sectors, forming a four-dimensional feature vector, which describes the distribution characteristics of the yaw impact in different directions.

[0069] The non-linear regression analysis of the yaw impact vector intensity is realized by using the support vector regression method. First, the original yaw impact vector intensity data is divided into a test set and a training set according to a ratio of 2:8. The training set contains 7000 records, and the test set contains 1745 records. Each record contains a yaw impact vector intensity value and its four sector projection components. The system performs standardization preprocessing on the training data. The Z-score standardization method is used to make the data mean value 0 and the standard deviation 1. The standardization parameters are saved for subsequent test data processing. The system constructs a support vector regression model. The kernel function is selected as the radial basis function (RBF). The hyperparameters are determined by grid search and 5-fold cross-validation. The penalty parameter C is set to 10, the kernel parameter γ is set to 0.05, and ε is set to 0.01. The iterative training process uses the sequential minimal optimization algorithm. The maximum number of iterations is set to 1000, and the convergence threshold is set to 0.001. After the model is trained, the performance is evaluated on the test set. The mean absolute error is 2.15, the root mean square error is 3.42, and the determination coefficient R² is 0.924. The system uses the trained model to regress and fit the yaw impact vector intensity data to generate the yaw impact vector regression intensity. The regression intensity is smoother, reducing the noise and fluctuations in the original data. The system calculates the braking disorder vector intensity in the time dimension based on the yaw impact vector regression intensity. First, a dynamic response model of the braking system is established. The model parameters include a braking actuator time constant of 25 ms, a braking pressure response delay of 15 ms, and a braking build-up time of 80 ms. The system uses a time series prediction method to estimate the trend of the yaw impact vector intensity in the next 100 ms. The prediction uses an autoregressive moving average model with an order of p=3 and q=2. The predicted result is combined with the current regression intensity to obtain the predictive yaw impact vector intensity. The system calculates the braking disorder vector intensity based on the current state of the vehicle and the predictive yaw impact vector intensity. Four correction factors are considered in the calculation: a vehicle speed correction factor, a road adhesion correction factor, a vehicle side slip correction factor, and a steering state correction factor. The vehicle speed correction factor ranges from 0.8 to 1.2 and increases with increasing vehicle speed. The road adhesion correction factor ranges from 0.7 to 1.4 and increases with decreasing adhesion coefficient. The vehicle side slip correction factor ranges from 0.9 to 1.3 and increases with increasing side slip angle. The steering state correction factor ranges from 0.85 to 1.15 and is related to the steering wheel angular velocity. The four correction factors are combined through multiplication. The system maps the corrected predictive yaw impact vector intensity to the braking disorder vector intensity using a piecewise linear function. The braking disorder vector intensity has a dimension of N·m·s and represents the moment impulse that needs to be intervened by the vehicle braking system. The value range is 0-3000. The system updates the braking disorder vector intensity value at a control period of 20 ms to ensure control real-time and stability. The final braking disorder vector intensity is used as a key input for subsequent braking control.

[0070] Step S234 includes the following steps:

[0071] According to the steering angle mutation swing amplitude, the yaw trajectory direction is analyzed, and the non-linear time sequence integral of the speed increasing proportion difference is performed to obtain acceleration speed increasing integral data;

[0072] The dynamic offset angle of the steering angle mutation swing amplitude is calculated, and the dynamic offset angle is generated;

[0073] According to the acceleration speed increasing integral data and the dynamic offset angle, the rotational inertia fluctuation estimation is performed, and then the angular momentum partial derivation calculation is performed to obtain rotational angular momentum partial derivation data;

[0074] Based on the acceleration speed increasing integral data, the dynamic offset angle and the rotational angular momentum partial derivation data, the oblique combined vector strength coupling analysis of the yaw trajectory direction is performed to obtain oblique combined vector strength coupling data; wherein the oblique combined vector is used to describe the motion or force of the vehicle in the longitudinal and transverse directions, and simultaneously contains the strength information of the two directions;

[0075] According to the oblique combined vector strength coupling data, the yaw impact vector strength is analyzed to obtain the yaw impact vector strength.

[0076] In the embodiment of the application, according to the steering angle mutation swing amplitude, the yaw trajectory direction is analyzed, first, the four steering angle mutation swing amplitude values output by step S233 are extracted, the unit is degree, each value corresponds to a 2s time window, the sign determination is performed on each swing amplitude value, if the positive cumulative change amount of the original steering wheel angle sequence is greater than the negative cumulative change amount in the window, the trajectory direction is marked as the right bias mark value +1, otherwise, the left bias mark value -1, at the same time, the maximum instantaneous angular velocity in the window is calculated, the unit is degree / s, as the direction strength weight factor, the direction sign and the strength weight are multiplied to obtain the signed yaw trajectory direction scalar value, a total of four values, which are used for subsequent coupling operation, and the non-linear time sequence integral operation of the speed increasing proportion difference sequence is performed, the input is the four speed increasing proportion difference amplitude values output by step S232, the unit is m / s 4, the trapezoidal numerical integration method is used to take the average of adjacent two points at a time step of 0.5s, multiply by 0.5s, and accumulate to obtain acceleration acceleration integral data, unit m / s³, the integral starting point is set to 0, and the terminal point is the 4th point, a total of 4 integral values are output, corresponding to the longitudinal dynamic energy accumulation of 4 time segments. The dynamic offset angle calculation is performed on the sudden change of the steering angle. First, the extreme points in each 2s window are extracted from the original steering angle sequence, including the local maximum and minimum, the absolute value of the angle difference between adjacent extreme points is calculated, if the difference is greater than 45° and the sign alternates, it is counted as a sudden change of direction, the number of sudden changes of direction in each window is counted, if the number is greater than or equal to 3, the dynamic offset angle calculation is started, the dynamic offset angle is equal to the average of all sudden change angles in the window multiplied by the attenuation coefficient 0.85, unit degree, if the number of sudden changes is less than 3, the dynamic offset angle is set to 0, and 4 dynamic offset angle values are output, each value is accurate to 1 decimal place in the range of 0 to 90°.

[0077] According to the acceleration acceleration integral data and the dynamic offset angle, the moment of inertia fluctuation is estimated. First, a simplified rigid body model of the vehicle is constructed, the mass is 1560kg, the mass center height is 0.52m, the wheelbase is 2.7m, the front wheelbase is 1.56m, and the rear wheelbase is 1.54m. The equivalent lateral displacement is calculated according to the dynamic offset angle, the formula is the mass center height multiplied by the angle radian value, the lateral offset distance is obtained, unit m, the instantaneous lateral force is calculated combined with the acceleration acceleration integral data, unit N, equal to the mass multiplied by the lateral acceleration, the lateral acceleration is obtained by the second order difference of the lateral offset distance to time, the sampling rate is 2Hz, the torque around the Z axis is calculated, unit N·m, equal to the lateral force multiplied by the mass center height, the instantaneous moment of inertia is estimated according to the relationship between the torque and the angular acceleration, unit kg·m², equal to the torque divided by the angular acceleration obtained by the difference of the yaw rate, unit rad / s², an estimated value of the moment of inertia is output for each time window, a total of 4 values, then the angular momentum partial derivative calculation method is performed on the angular momentum values of adjacent time windows, unit kg·m² / s, the first order central difference is performed to obtain the angular momentum change rate, i.e. the angular momentum partial derivative data, unit kg·m² / s², a total of 3 output values, corresponding to the transition segments of the 1st to 2nd, 2nd to 3rd, and 3rd to 4th time windows.

[0078] Based on the acceleration speed integral data dynamic offset angle and the rotation angle momentum offset data, the lateral trajectory azimuth is analyzed by oblique vector intensity coupling. Firstly, the acceleration speed integral data is taken as the longitudinal component, with the unit of m / s³. The dynamic offset angle is converted into radian, multiplied by the whole vehicle mass 1560 kg, and then multiplied by the gravity acceleration 9.8 m / s² to obtain the transverse component, with the unit of N. The rotation angle momentum offset data is taken as the coupling weight factor, with the unit of kg·m² / s². The three weighted synthesis operations are performed. The longitudinal component is multiplied by the weight factor to obtain the modified longitudinal intensity. The transverse component is multiplied by the weight factor to obtain the modified transverse intensity. The modified longitudinal intensity and the modified transverse intensity are combined into a two-dimensional vector. The oblique vector intensity coupling data is obtained by calculating the Euclidean norm, with the unit of N·m / s³. The same operation is independently performed on all 3 groups of transition section data to output 3 oblique vector intensity coupling values. Each value is kept to 4 digits after the decimal point, and is processed without normalization, standardization and dimensionless. The lateral impact vector intensity is analyzed according to the oblique vector intensity coupling data. The method is to perform sliding maximum value filtering on the 3 oblique vector intensity coupling values, with the window length of 2 and the step length of 1. 2 peak intensity values are output. The larger value is multiplied by the time integral factor 2s to obtain the energy equivalent value, with the unit of N·m·s. The energy equivalent value is divided by the tire ground print effective length 0.22m to obtain the equivalent lateral impact force, with the unit of N. Finally, the equivalent lateral impact force is multiplied by the force arm coefficient 0.52m, i.e. the height of the center of mass, to obtain the lateral impact vector intensity, with the unit of N·m. The final scalar value is output to 3 digits after the decimal point, which is used for regression analysis in step S235.

[0079] Step S3 includes the following steps:

[0080] Step S31: The unordered vector segment clustering data is processed by convolution to obtain unordered vector segment convolution data.

[0081] Step S32: Brake timing intervention correction perception is performed according to the unordered vector segment convolution data, so as to obtain brake intervention correction data.

[0082] Step S33: The brake intervention correction data is iteratively learned to generate brake intervention correction optimization data.

[0083] Step S34: The brake intervention correction optimization data is sent to the new energy vehicle control terminal to perform new energy vehicle safe driving control.

[0084] As an example of the present application, refer to Figure 3 In this example, the step S3 includes:

[0085] Step S31: The unordered vector segment clustering data is processed by convolution to obtain unordered vector segment convolution data.

[0086] In the embodiment of the application, the convolution processing of the unordered vector segment clustering data is realized by using a multi-channel time sequence convolution network technology. First, the unordered vector segment clustering data is organized into a standard time sequence data structure, and the data dimension is [sample number x time step x feature number], that is, [8745 x 400 x 16], wherein the sample number is the total number of emergency escape events, the time step corresponds to the number of sampling points in a 2-second sampling window, and the feature number includes the yaw impact vector intensity value, the clustering label, the confidence and other key parameters. The system performs normalization preprocessing on the time sequence data, adopts the Z-score standardization method, so that the mean value of each feature is 0 and the standard deviation is 1, the standardization parameters are calculated by batch statistics, and the normalized data is input into a multi-channel convolution processing module. The module adopts a hierarchical convolution structure and includes three convolution layers. The first layer convolution uses 32 filters, the kernel size is 3 x 1, the step is 1, the padding mode is SAME, and the activation function is ReLU. The second layer convolution uses 64 filters, the kernel size is 5 x 1, the step is 2, the padding mode is SAME, and the activation function is ReLU. The third layer convolution uses 128 filters, the kernel size is 7 x 1, the step is 2, the padding mode is SAME, and the activation function is ReLU. Batch normalization processing is performed after each layer of convolution, and the parameter settings are as follows: the momentum is 0.99, the epsilon is 0.001, and the decay rate is 0.9. The system applies residual connection between the convolution layers to improve the gradient propagation efficiency, and the residual block adopts identity mapping. Parallel computing of the convolution processing is realized on an NVIDIA Jetson AGX Xavier platform, the number of CUDA cores is 512, the computing performance reaches 32 TOPS, the processing delay is controlled within 5 ms, the system performs channel attention mechanism processing on the feature map output by the third layer of convolution, adopts a Squeeze-and-Excitation module, and the compression ratio is 16. The mechanism adaptively adjusts the weights of different channels, highlights important features and suppresses secondary features. The feature map after channel attention processing applies spatial attention mechanism, adopts a self-attention algorithm, the number of heads is set to 8, the hidden layer dimension is 64, and the dropout rate is 0.1. The spatial attention highlights the key time period in the time sequence data. The system fuses the feature maps after channel attention and spatial attention processing, adopts weighted summation as the fusion method, the weights are automatically learned through back propagation, the fused features are subjected to global average pooling and global maximum pooling, the pooled features are spliced to form a compact representation, and finally, the features are reduced in dimension through 1 x 1 convolution, the number of filters is 64, and unordered vector segment convolution data is formed, the data dimension is [8745 x 100 x 64], wherein the time step is compressed to 1 / 4 of the original, and the feature dimension is expanded to 4 times of the original. The convolution data contains the key patterns and time sequence features of the original clustering data. The system performs quality evaluation on the convolution data, calculates the signal-to-noise ratio, and the average signal-to-noise ratio is improved by 15.3 dB. The feature discrimination is evaluated by the contour coefficient and is improved from 0.68 of the original data to85, indicating that the convolution processing effectively extracts the key features in the disordered vector segment clustering data, while suppressing the noise interference, and the system delivers the disordered vector segment convolution data to the subsequent brake timing intervention correction perception module, to provide a basis for accurate intervention control.

[0087] Step S32: brake timing intervention correction perception is performed according to the disordered vector segment convolution data, so as to obtain brake intervention correction data.

[0088] In the embodiment of the application, the brake timing intervention correction perception according to the disordered vector segment convolution data is implemented by using a multi-stage decision fusion technology, first, the timing features are extracted from the disordered vector segment convolution data, a bidirectional gate recurrent unit network is used, the network structure includes 2 layers of BiGRU, the number of hidden units in each layer is 128, the activation function is tanh, the cycle dropout rate is 0.2, the input dropout rate is 0.3, the network training uses an Adam optimizer, the learning rate is 0.001, β1 is 0.9, β2 is 0.999, and ε is 1 , size is set to 64, training iterations are 100 rounds, the features extracted by BiGRU network contain long-term dependence of time series, the system extracts segmented direction offset angle and speed stall value of different time periods from the convolution data, the direction offset angle is recovered from the convolution features through deconvolution algorithm, the deconvolution uses transpose convolution layer, the kernel size is 4, the step is 2, and the channel number is 16, the speed stall value is also recovered through deconvolution, the parameter settings are the same, and the recovery accuracy is more than 95% of the original data, the system quantifies the body offset angle based on the recovered direction offset angle, adopts adaptive quantization algorithm, and divides the continuous angle value into 7 discrete levels: extremely small offset (0-3 degrees), small offset (3-8 degrees), small-medium offset (8-15 degrees), medium offset (15-25 degrees), medium-large offset (25-35 degrees), large offset (35-50 degrees) and extremely large offset (more than 50 degrees), the quantization threshold is automatically determined through K-means clustering to ensure that the intra-class distance is minimum and the inter-class distance is maximum, the system analyzes the braking torque demand based on the body offset angle and the speed stall value, adopts fuzzy logic controller, uses triangular and trapezoidal membership functions for input fuzzification, divides the angle input into 7 fuzzy sets, divides the speed stall value into 5 fuzzy sets, and divides the output fuzzy set into 9 levels. Fuzzy rule base contains 35 IF-THEN rules, reasoning method uses Mamdani maximum minimum algorithm, defuzzification uses centroid method, and the output range of fuzzy controller is 0-3000 N·m, corresponding to actual braking torque demand. The system performs time series smoothing processing on the braking torque demand based on the direction offset angle and the speed stall value, adopts nonlinear exponential moving average algorithm, and the smoothing factor a is adaptively adjusted according to the offset angle change rate. When the change rate is large, the value of a is small (0.2-0.4), and when the change rate is small, the value of a is large (0.6-0.8), so as to ensure rapid response in severe working conditions and suppress fluctuations in stable working conditions. The system inputs the smoothed braking torque demand sequence into the time series decision tree integrator, the integrator contains 5 decision trees, the tree depth is 8, the minimum leaf node sample number is 15, the splitting standard is Gini impurity, and the integration method adopts weighted voting, and the weight is optimized through gradient boosting. The output of the decision tree integrator is the braking intervention time and the intensity recommendation value, the system performs multi-source fusion based on the time series decision and the fuzzy control result, adopts Dempster-Shafer evidence theory for fusion, defines three focus elements: need braking, observation, and no need braking, calculates the basic probability distribution of each source, calculates the fusion decision credibility through the DS combination rule, triggers the braking intervention when the "need braking" focus element credibility exceeds 0.85, the intervention intensity is based on the fuzzy controller output, and the intervention time is based on the time series decision tree output. The system generates braking intervention correction data by comprehensively considering various decision sources, and the data format includes key parameters such as intervention time point, intervention duration, intensity curve and direction correction suggestion, which provides input for subsequent iterative optimization.

[0089] Step S33: Iterative learning on the brake intervention correction data to generate brake intervention correction optimization data;

[0090] In the embodiment of the application, the hierarchical reinforcement learning framework is used to implement the iterative learning of the brake intervention correction data. First, a brake intervention state space is constructed, the state vector dimension of which is 18, including 10 vehicle dynamic state parameters (vehicle speed, acceleration, yaw rate, side slip angle, etc.) and 8 brake system state parameters (brake pressure, brake distribution ratio, wheel speed difference, etc.). The state space is represented in a discrete-continuous mixed manner, the action space is defined as a two-dimensional continuous space, representing the brake torque size (0-3000 N·m) and the brake force distribution ratio (0.3-0.7) respectively, the reward function is designed considering safety, comfort and handling, which respectively account for 50%, 30% and 20% of the weight, the safety reward is based on the risk avoidance success rate, the comfort reward is based on the passenger body acceleration smoothness, and the handling reward is based on the vehicle response sensitivity. The system uses a hierarchical reinforcement learning structure, the high-level strategy is responsible for decision-making brake intervention timing, and the low-level strategy is responsible for executing brake torque control. The high-level uses the deep Q network (DQN) algorithm, the network structure is a fully connected neural network, the number of layers is 4, the number of hidden layer nodes is [128, 256, 128, 64], the activation function is LeakyReLU, α is 0.01, the experience replay buffer size is 10000, the batch size is 32, the discount factor γ is 0.95, the target network update frequency is 100 steps, the low-level uses the deep deterministic policy gradient (DDPG) algorithm, the Actor network structure is [128, 256, 128, 64], the Critic network structure is [256, 512, 256, 128], both networks use BatchNorm and Dropout regularization, the Actor learning rate is 0.0001, the Critic learning rate is 0.001, the target network soft update parameter τ is 0.001, the noise uses the Ornstein-Uhlenbeck process, θ is 0.15, and σ is 0.2, The system uses simulation environment for reinforcement learning training, the simulation environment is based on CarMaker professional vehicle dynamics simulation platform, contains 16 typical risk avoidance scenes and 4 road adhesion conditions, the simulation sampling frequency is 1000Hz, the control frequency is 100Hz, the system carries out distributed training for different working conditions, uses asynchronous advantage Actor-Critic(A3C) framework, the number of parallel environments is 16, the number of working threads is 8, the number of training iterations is 5 million steps, the policy evaluation is carried out every 250000 steps, the evaluation indexes include average return, success rate and control stability, the training process uses curriculum learning strategy, gradually increases the difficulty from simple scene, the system applies the trained strategy to brake intervention correction data for optimization, uses iterative optimization process, each iteration includes prediction, evaluation and adjustment three steps, the prediction step uses reinforcement learning strategy to generate candidate brake intervention scheme, the evaluation step calculates the expected return of the scheme, the adjustment step adjusts the strategy parameters according to the evaluation result, the number of iterations is set to 5 times, the convergence standard is that the difference between the schemes of adjacent two iterations is less than 3%, the system optimizes the brake intervention time, force curve and duration three key parameters through iterative learning, for each brake intervention event, the system considers multiple alternative schemes, selects the scheme with the highest expected return as the final optimization result, generates brake intervention correction optimization data, the data format uses structured binary storage, contains control parameters, timing curve and metadata three parts, the total data amount is 8745 optimization records, each record size is about 24KB, the total amount is about 210MB, the system compresses the optimization data, uses lossless compression algorithm, the compression ratio reaches 3:1, the finally generated optimization data is used for actual execution of vehicle control system.

[0091] Step S34: Send the brake intervention correction optimization data to the new energy vehicle control terminal to execute the new energy vehicle safe driving control.

[0092] In the embodiment of the application, the brake intervention correction optimization data is sent to the new energy vehicle control terminal by using a multi-level secure transmission architecture, first, the brake intervention correction optimization data is encrypted, the data is encrypted by using an AES-256 algorithm, the key management adopts an elliptic curve cryptography system, the curve is selected as NIST P-256, the key length is 256 bits, the digital signature adopts an ECDSA algorithm, the data integrity and source authentication are ensured, the encrypted data is transmitted through a vehicle-mounted Ethernet, an IEEE802.3bw 100BASE-T1 standard is adopted, the bandwidth is 100 Mbps, the physical layer adopts a single pair of twisted pair, the transmission distance reaches 15 m, the network topology adopts a star structure, the central switch is a vehicle-mounted level, supports a time-sensitive network (TSN) technology, guarantees the real-time performance of the key data, the data transmission adopts a layered protocol stack, the physical layer and the data link layer adopt the vehicle-mounted Ethernet, the network layer adopts IPv6, the transmission layer adopts a mixed mechanism of UDP and TCP, the key real-time control data is transmitted through UDP, the configuration and management data are transmitted through TCP, the application layer adopts a customized protocol, the packet header length is 32 bytes, and the information includes the data type, the priority, the timestamp and the sequence number, the system implements a multi-level redundancy mechanism in the transmission process, the key data is transmitted through two physical channels at the same time, a double redundancy fault-tolerant technology is adopted, the system can still work normally when a single point fault occurs, the data is authenticated and the integrity is checked after reaching the new energy vehicle control terminal, the authentication adopts a challenge-response mechanism, the checking adopts an HMAC-SHA256 algorithm, after the checking is passed, the data is decrypted and loaded into the memory of the control terminal, the control terminal hardware platform adopts a security domain isolation architecture, the main processor is an NXPS32G automobile processor, the main frequency is 2.0 GHz, the core is 8, the memory is 4 GB LPDDR4, the storage is 32 GB eMMC, and a hardware security module HSM is arranged, supports secure boot, real-time encryption and decryption and key management, the decrypted brake intervention correction optimization data is distributed to three execution controllers by the system: a brake control unit (BCU), a motor control unit (MCU) and a vehicle body stability control unit (ESC), the distribution adopts a controller area network CANFD, the baud rate is 5 Mbps, the frame format adopts an extended frame, the priority is based on identifier arbitration, the BCU receives the brake force distribution and pressure control data, and precise brake torque control is implemented, and the control resolution is 0.1MPa, response time less than 15ms, MCU receives motor torque modulation data, realizes cooperative control of regenerative braking and hydraulic braking, torque control precision reaches ±2%, response time is less than 10ms, ESC receives yaw stability control data, executes differential braking control, stability control frequency is 100Hz, the system performs final safety check before executing control, the check items include data timeliness, control quantity rationality and system state consistency, after passing the check, the braking intervention control is performed according to the predetermined time sequence, the control process is divided into three stages: pre-braking stage (slight braking force is established, and lasts for 100-200ms), main braking stage (braking force is applied according to the force curve, and lasts for 300-1500ms) and exit stage (braking force is smoothly reduced, and lasts for 200-400ms), smooth transition is realized between each stage through slope limitation, the slope range is 5-20MPa / s, the system continuously monitors the vehicle state and the driver's response during the execution process, when the driver initiates the takeover (such as stepping on the brake or turning the steering wheel sharply), the automatic control is immediately smoothly exited, the exit slope is 30MPa / s, ensuring smooth handover of control right, the system records the execution results to the event log, including trigger conditions, control parameters, vehicle response and termination reasons and the like, the log data is uploaded to the cloud server through the 4G / 5G network, for subsequent analysis and system optimization, and the whole process of new energy vehicle safe driving control is completed.

[0093] Step S32 includes the following steps:

[0094] Step S321: Extract the segmented direction offset angle and vehicle speed stall value of different time periods in the disordered vector segmented convolution data.

[0095] Step S322: Quantize the body offset angle according to the segmented direction offset angle and vehicle speed stall value of different time periods, and obtain the body offset angle; perform braking torque demand analysis based on the body offset angle, and obtain braking torque demand data;

[0096] Step S323: Based on the segmented direction offset angle and vehicle speed stall value of different time periods, the braking torque demand data is subjected to braking force smoothing control optimization in different time periods, and braking force smoothing control data is obtained.

[0097] Step S324: According to the braking force smoothing control data, the braking timing intervention correction perception is performed, so as to obtain braking intervention correction data.

[0098] In the embodiment of the present application, the segmented direction offset angle and the vehicle speed stall value of different time periods in the unordered vector segmented convolution data are extracted by using a deconvolution reconstruction technology. First, the unordered vector segmented convolution data is structured and analyzed. The convolution data has a dimension of [8745x100x64] and contains time compression and feature expansion information. A transposed convolution network is used for feature recovery. The network includes three transposed convolution layers. The first layer has the following parameters: filter number 32, kernel size 7x1, step size 2, padding mode SAME, and activation function ReLU. The second layer has the following parameters: filter number 16, kernel size 5x1, step size 2, padding mode SAME, and activation function ReLU. The third layer has the following parameters: filter number 8, kernel size 3x1, step size 1, padding mode SAME, and activation function linear. Each transposed convolution layer is followed by a batch normalization layer with the following parameters: momentum 0.9 and epsilon 0.001. The transposed convolution network outputs a dimension of [8745x400x8], which recovers the time series data with the original sampling rate. The segmented direction offset angle information is extracted from the recovered data by using a frequency domain feature back projection technology. First, the recovered data is subjected to a short-time Fourier transform with a window size of 128 points, an overlap rate of 75%, and a window function of Hanning. A time-frequency spectrum is obtained, and the frequency resolution is 1.56 Hz. The angle-related frequency band components are extracted from the time-frequency spectrum, mainly in the frequency band of 0.5-4 Hz. The frequency band information is converted back to the time domain by inverse Fourier transform to obtain the direction offset angle time series with a sampling rate of 100 Hz and a resolution of 0.1 degrees. The system divides a 2-second time window into 5 time periods, each with a length of 400 ms, corresponding to the preparation stage (0-400 ms), the initial response stage (400-800 ms), the mid-response stage (800-1200 ms), the stable stage (1200-1600 ms), and the recovery stage (1600-2000 ms). Statistical features are calculated for the direction offset angle data in each time period, including the average angle, the maximum angle, the angle change rate, and the angle fluctuation amplitude, to form the angle feature vector of each time period. The vehicle speed stall value is extracted from the recovered data by using a signal separation technology. First, the independent component analysis (ICA) algorithm is used to separate the mixed signal into independent components with the following parameters: maximum iteration number 200 and tolerance 0.0001. Among the separated independent components, the vehicle speed-related component is identified by spectral feature, and the vehicle speed signal feature frequency is usually lower than 0.5 Hz. The extracted vehicle speed signal is smoothed by using a Kalman filter with the following filter parameters: process noise covariance 0.01 and measurement noise covariance 0.1. The vehicle speed stall value is calculated as the difference between the actual vehicle speed and the expected safe vehicle speed, which is calculated by a curvature safety speed model with the following model parameters: road curvature, lateral acceleration threshold (set to 3.5 m / s²), and safety margin coefficient (set to 0.85), the system also segments the vehicle speed stall values according to the same 5 time periods, calculates the average stall value, maximum stall value, stall rate of change and stall trend in each segment, forms the stall feature vector of each time period, and combines the segmented directional offset angle and vehicle speed stall value to form a multi-dimensional feature tensor with dimensions [8745x5x16], where 8745 is the number of samples, 5 is the number of time periods, and 16 is the number of features in each time period. The feature tensor is stored in HDF5 format, supporting efficient random access. The extracted feature data is subjected to integrity verification and outlier detection to ensure data quality. The integrity check includes missing value detection and continuity verification. The outlier detection uses a modified Z-score method with a threshold of 3.5, and the proportion of outliers is controlled within 0.3% of the total data to ensure the quality of the data for subsequent analysis.

[0099] The quantification of the body offset angle according to the segmented directional offset angle and vehicle speed stall value in different time periods is achieved using an adaptive clustering quantization technique. The system first performs time weighting on the segmented directional offset angle, with the weighting function being an exponential decay function and the weight coefficient being where is the attenuation coefficient, set as 0.5, t is the time offset relative to the current time, in seconds, the weighting process highlights the importance of recent angle changes, the system applies a Gaussian Mixture Model to the weighted angle data for probability density estimation, the model parameters are set as 5 components, covariance type as full, initialization method as k-means++, maximum number of iterations as 200, convergence threshold as 0.001, the GMM model fitted by the EM algorithm captures the multi-modal characteristics of the angle distribution, the system designs a quantization threshold based on the GMM model, determines the optimal threshold using the Bayesian Information Criterion (BIC) method, realizes adaptive quantization of the direction offset angle, the quantization level is set as 7 discrete levels: very small offset (0-2.5 degrees), small offset (2.5-7.5 degrees), small-medium offset (7.5-15 degrees), medium offset (15-25 degrees), medium-large offset (25-37.5 degrees), large offset (37.5-55 degrees) and very large offset (more than 55 degrees), similar quantization is performed on the speed stall value, the quantization level is set as 5 discrete levels: very small stall (0-5 km / h), slight stall (5-15 km / h), medium stall (15-30 km / h), severe stall (30-50 km / h) and extreme stall (more than 50 km / h), the system calculates the vehicle body offset angle based on the quantized direction offset angle and speed stall level, combined with the current state parameters of the vehicle (including yaw rate, lateral acceleration and tire side slip angle), the calculation method uses a nonlinear mapping network, the network structure is a fully connected feedforward network, the layer structure is [16, 32, 16, 8, 1], the activation function is Leaky ReLU, alpha is 0.1, the output layer activation function is tanh, the network training uses the mean square error loss function, the optimizer is Adam, the learning rate is 0.001, beta1 is 0.9, beta2 is 0.999, the training data is 12000 expert-labeled samples, the validation accuracy reaches an average absolute error of 0.85 degrees, the quantized value of the vehicle body offset angle ranges from -45 degrees to +45 degrees, a positive value indicates a right deviation, a negative value indicates a left deviation, the brake torque demand analysis based on the vehicle body offset angle uses a multi-objective optimization method, first, a vehicle yaw dynamics model is established, the model parameters include vehicle mass 1850 kg, wheelbase 2.85 m, front and rear wheel track 1.62 m and 1.65 m respectively, center of mass height 0.58 m, front and rear wheel cornering stiffness 95000 N / rad and 105000 N / rad respectively, the system calculates the yaw moment required to stabilize the vehicle body based on the dynamics model, taking into account the sideslip angle, yaw rate and speed, for a given vehicle body offset angle, the system calculates the yaw moment required to restore the neutral state, converts it to an equivalent brake torque demand, the conversion coefficient is determined according to the vehicle geometry and mass characteristics, the typical value is 1.2-1.8. The system performs multi-objective optimization on braking torque demand, optimization objectives include minimizing steady time, minimizing lateral offset distance and minimizing ride comfort impact, optimization algorithm adopts sequential quadratic programming (SQP) method, constraint conditions include maximum braking torque limit (3500 N·m), braking torque rate limit (800 N·m / s) and braking force distribution limit (front and rear wheel braking proportion range 0.4-0.7), optimization solution adopts augmented Lagrange multiplier method, iteration number is set to 50, convergence threshold is 0.001, the system generates braking torque demand data according to optimization results, data includes total braking torque, front and rear wheel distribution proportion, left and right wheel distribution proportion and time sequence curve, torque range is 0-3500 N·m, time resolution is 10 ms, curve adopts piecewise cubic Hermite interpolation to express, ensuring first derivative continuity and smooth transition.

[0100] The braking force torque demand data is optimized for different time periods based on segmented direction deviation angles, vehicle speed stall values, and braking force torque demand data for different time periods. The system first subdivides the 2-second control window into 8 time periods, each 250 ms, and designs different control strategies for different time periods. For the direction deviation angle and vehicle speed stall value of each time period, the system calculates the direction deviation intervention correction angle and intervention correction speed. The calculation method uses an adaptive PID controller, and the PID parameters are optimized offline by a particle swarm optimization algorithm. The optimization goal is to minimize the overshoot and settling time. For the direction deviation intervention correction angle, the PID parameters are set to Kp=0.8-1.2 (adaptive according to speed), Ki=0.05-0.15, and Kd=0.2-0.4. For the intervention correction speed, the PID parameters are set to Kp=0.4-0.6, Ki=0.03-0.08, and Kd=0.1-0.2. The system simulates the vehicle intervention correction braking jitter deviation based on the intervention correction angle and correction speed. A high-precision vehicle dynamics simulation model is used, which includes 14 degrees of freedom, considers suspension geometry, tire nonlinear characteristics, and steering system characteristics, and has a simulation step size of 1 ms. The simulation model parameters are calibrated through real vehicle testing, including suspension stiffness, damping characteristics, steering transmission ratio, and tire Magic Formula parameters. The simulation calculates the vehicle body response under different braking force torque inputs, including yaw rate, lateral acceleration, and side slip angle. The system analyzes the simulation results to calculate the jitter deviation index during braking. The index is defined as the energy proportion of the high-frequency component (5-15 Hz) of the yaw rate. In general, an index below 5% indicates good smoothness, and an index above 15% indicates significant jitter. The system performs multiple regression analysis on the jitter deviation data using a support vector regression algorithm with a radial basis function (RBF) kernel function, C=10, γ=0.1, and ε=0.01. The regression model inputs are braking force torque, braking force change rate, and road adhesion coefficient, and the output is the predicted jitter deviation index. The root mean square error of the regression model is 1.8%. The system gradually controls and optimizes the direction deviation intervention correction angle and intervention correction speed based on the jitter deviation regression data. The optimization uses a model predictive control (MPC) framework with a prediction horizon of 500 ms, a control horizon of 100 ms, and a sampling period of 20 ms. The state variables include body slip angle, yaw rate, longitudinal speed, and lateral speed. The control variables are braking force torque and braking force distribution ratio. The system state equation is based on a linearized vehicle dynamics model, and the control objective function is in the form of weighted square sum. The weight matrix is determined through multi-objective optimization. The constraint conditions for control input include braking force torque range [0, 3500] N·m, torque change rate range [-800, 800] N·m / s, and braking force distribution range [0.4, 0.7], the optimization problem is solved in real time by a quadratic programming solver, the solver uses the active set method, the maximum number of iterations is 50, the calculation time is controlled within 5 ms, the system generates offset progressive correction angle and progressive correction speed according to the MPC optimization result, the progressive correction angle change rate is limited to 15 degrees per second, and the progressive correction speed change rate is limited to 5 km / h per second, which ensures the smooth change of the control quantity, the system performs brake force smooth control optimization based on the offset progressive correction angle and the progressive correction speed, adopts a nonlinear filter combination control strategy, the filter includes a low-pass filter, a slope limiter and a jitter suppressor, the low-pass filter cutoff frequency is 5 Hz, the slope limiter parameter is set to the rising slope 500 N·m / s, the falling slope 600 N·m / s, the jitter suppressor uses an adaptive threshold method, the threshold range is 50-200 N·m, which is dynamically adjusted according to the vehicle speed and the road adhesion coefficient, the brake force smooth control data generated by the system includes brake torque time sequence, brake force distribution parameters and control timing markers, the data sampling interval is 10 ms, the duration is 2 seconds, forming a 200-point control sequence.

[0101] The braking timing intervention correction awareness according to the braking force smoothing control data is achieved by using a multi-stage decision fusion method. The system first performs time series analysis on the braking force smoothing control data, extracts key event points by using a sliding window method, the window size is 150 ms, the step size is 30 ms, identifies time points where the braking torque rate exceeds the critical value, the critical value is set to 250 N·m / s, these time points are used as potential intervention opportunity candidate points, the system calculates the intervention suitability score for each candidate point, the score considers four factors: vehicle stability margin, driver operation state, road adhesion condition and expected effect, the weight of each factor is 0.4, 0.3, 0.2 and 0.1 respectively, the stability margin is calculated by phase space analysis, the vehicle yaw rate and sideslip angle are mapped to a two-dimensional phase plane, the distance from the current state to the instability boundary is calculated, the driver operation state is judged by the steering wheel angle, the steering wheel angle speed and the pedal operation, the road adhesion condition is estimated by the wheel speed difference and the acceleration response, the expected effect is calculated by simulation prediction, the system uses a Bayesian decision network to make intervention timing decision, the network structure includes three layers: evidence layer, reasoning layer and decision layer, the input of the evidence layer is the vehicle state parameters and the braking force smoothing control data, the reasoning layer calculates the conditional probability of different intervention timing, the decision layer selects the intervention timing with the maximum expected utility, the network parameters are trained by historical data, the number of training samples is 15000, the verification accuracy rate reaches 93.5%, the system determines the optimal intervention timing, calculates the accurate braking intervention timing, the timing design includes three parts of pre-braking stage, main braking stage and exit stage, the pre-braking stage lasts for 80-120ms, the braking torque slope is 150N·m / s, the main braking stage lasts for 400-1200ms, the braking torque changes according to the optimization curve, the exit stage lasts for 150-250ms, the braking torque slope is-200N·m / s, the system calculates the direction correction auxiliary quantity in the braking force intervention process, the calculation is based on the vehicle yaw dynamics model and the steering system characteristics, the direction correction includes active steering compensation and torque superposition, the active steering compensation range is ±3 degrees, the steering torque superposition range is ±5N·m, the correction quantity is dynamically adjusted during braking, the system further optimizes the transition characteristics of braking intervention, adopts state feedback and feedforward combined control method, the feedback control is based on the body sideslip angle and yaw rate, the feedforward control is based on the steering wheel angle and angular velocity, the controller parameters are optimized by H∞ robust control theory to ensure the stability under parameter uncertainty, the system encapsulates the optimized braking intervention correction data, the data format includes four parts: control instruction header, timing parameter block, control quantity data block and state monitoring block, the control instruction header includes instruction type, priority, sequence number and check code, the timing parameter block includes the duration of each stage, trigger time and exit condition, the control quantity data block includes braking torque sequence, distribution ratio and direction correction quantity, the state monitoring block includes key state variable threshold and emergency handling strategy, the system performs integrity verification and timeliness marking on the braking intervention correction data, the integrity verification adopts CRC32 check algorithm, the timeliness marking includes generation timestamp and validity period, the data transmission adopts real-time transport protocol, the priority is set to the highest to ensure the timely delivery of control instructions, the braking intervention correction data generated by the system realizes accurate control of braking intervention timing, intensity and duration in the emergency avoidance scene, forms a closed-loop control system, and provides a core decision basis for new energy vehicle safety driving control system.

[0102] Step S323 includes the following steps:

[0103] Based on the segmented direction deviation angle, the vehicle speed stall value, the direction deviation intervention correction angle and the intervention correction vehicle speed in different time periods are determined;

[0104] According to the direction deviation intervention correction angle and the intervention correction vehicle speed, the vehicle intervention correction braking jitter deviation simulation calculation of braking torque demand data in different time periods is performed to obtain jitter deviation data;

[0105] The multiple regression analysis is performed on the jitter deviation data to generate jitter deviation regression data;

[0106] The direction deviation progressive correction angle and the intervention correction vehicle speed of different time periods are obtained based on the jitter deviation regression data.

[0107] The brake force smooth control data is obtained by performing brake force smooth control optimization of different time periods according to the deviation progressive correction angle and the progressive correction vehicle speed.

[0108] In the embodiment of the application, the direction deviation intervention correction angle and the intervention correction vehicle speed in different time periods are determined based on the segmented direction deviation angle, the vehicle speed stall value in different time periods, and are realized by using a multi-channel adaptive control technology. First, a 2-second control window is divided into 8 time periods according to dynamic response characteristics, i.e., a prediction period (0-200 ms), an initial response period (200-400 ms), a transition period I (400-600 ms), an enhancement period (600-900 ms), a stable period (900-1200 ms), a transition period II (1200-1500 ms), a weakening period (1500-1800 ms), and an exit period (1800-2000 ms). The corresponding direction deviation angle and vehicle speed stall value are extracted for each time period, a time-state matrix is constructed, the dimension of the matrix is 8x2, a nonlinear state converter is applied to the matrix data, the converter uses a segmented function mapping, the direction deviation angle is mapped to the direction deviation intervention correction angle, and the mapping rule is: when the deviation angle is less than 5 degrees, the correction angle = deviation angle x 0.2, when the deviation angle is between 5 and 15 degrees, the correction angle = 1 + deviation angle x 0.5, when the deviation angle is between 15 and 30 degrees, the correction angle = 6 + deviation angle x 0.3, and when the deviation angle is greater than 30 degrees, the correction angle = 10.5 + deviation angle x 0.15. According to experimental data statistics, the sensitivity coefficient of right deviation correction is 5-8% higher than that of left deviation correction, therefore, different parameters are set for right and left deviation conditions, the left deviation coefficient is the above value multiplied by 0.95, and the system applies a similar segmented function mapping to the vehicle speed stall value, maps the vehicle speed stall value to the intervention correction vehicle speed, and the mapping rule is: when the stall value is less than 5 km / h, the correction speed = stall value x 0.6, when the stall value is between 5 and 15 km / h, the correction speed = 3 + stall value x 0.8, when the stall value is between 15 and 30 km / h, the correction speed = 11 + stall value x 0.6, and when the stall value is greater than 30 km / h, the correction speed = 20 + stall value x 0.4. The unit of the mapped intervention correction vehicle speed is km / h, indicating a target vehicle speed value that needs to be reduced by braking intervention. The system considers the dynamic characteristics in different time periods, applies time weighting processing to the correction angle and the correction speed, and the weighting coefficient matrix is a preset 8x2 matrix, wherein the first column is the angle weighting coefficient, and the second column is the vehicle speed weighting coefficient. The angle weighting coefficient sequence is [0.4, 0.6, 0.8, 1.0, 0.9, 0.7, 0.5, 0.3], and the vehicle speed weighting coefficient sequence is [0.3, 0.5, 0.7, 0.9, 1.0, 0.8, 0.6, 0.4], the system applies smoothing processing to the weighted sequence, adopts Savitzky-Golay filtering algorithm, polynomial order is 3, window length is 5, ensures the smoothness of the sequence change, the system converts the processed correction angle and correction speed sequence into the standard format required by the control execution unit, including time stamp, correction absolute value and direction flag, angle accuracy is 0.1 degree, speed accuracy is 0.1 km / h, the system dynamically adjusts the correction according to the current speed, road adhesion coefficient and steering state, the adjustment coefficient is obtained by table lookup method, a three-dimensional lookup table is established, the size is 10x8x5, corresponding to 10 speed intervals (0-150 km / h, interval 15 km / h), 8 adhesion coefficient intervals (0.1-0.9, interval 0.1) and 5 steering states (sharp steering, moderate steering, slight steering, straight driving and reverse steering), the finally generated direction offset intervention correction angle and intervention correction speed data are used as input parameters for subsequent jitter yaw simulation.

[0109] The vehicle intervention correction braking shimmy yawing simulation calculation of the braking torque demand data in different time periods is realized by using high-precision vehicle dynamics simulation technology. The system first constructs a 14-degree-of-freedom vehicle dynamics model, including 6 body degrees of freedom (longitudinal, lateral, vertical, roll, pitch and yaw) and 8 wheel degrees of freedom (4 wheel rotation and 4 vertical displacement), the model parameters are determined based on the physical characteristics of a certain medium-sized SUV electric vehicle, including vehicle mass 1850kg, wheelbase 2.78m, front and rear wheel track 1.61m and 1.63m respectively, suspension stiffness front wheel 32000N / m, rear wheel 35000N / m, suspension damping front wheel 2800N·s / m, rear wheel 3000N·s / m, tire using Pacejka magic formula model, parameters B=10, C=1.9, D=1, E=0.97, model calculation using 4-order Runge-Kutta integration method, step length 1ms, ensure calculation accuracy, the system converts the direction offset intervention correction angle to steering system input, calculates the equivalent steering wheel angle through the steering transmission ratio (16:1), considers the elastic and clearance characteristics of the steering system, converts the intervention correction speed to the braking system input, calculates the equivalent brake pressure through the brake efficiency model (brake deceleration vs brake pressure relationship curve), the pressure range is 0-12MPa, the system constructs the braking torque distribution model, based on the vehicle load distribution and tire adhesion characteristics, calculates the four-wheel braking torque distribution ratio, the front and rear axle distribution ratio range is 60:40 to 50:50, the left and right wheel distribution ratio range is 45:55 to 55:45, according to the braking torque demand data and the braking force distribution ratio, the four-wheel braking torque time series is calculated, the braking torque of each wheel ranges from 0 to 1000N·m, the system simulates the dynamic characteristics of the brake actuator, including execution delay (15-25ms), pressure rise rate (5-10MPa / s) and pressure fluctuation (±0.3MPa), the brake actuator system uses a linear second-order model to represent, the natural frequency is 15Hz, the damping ratio is 0.7, the system simulates the braking shimmy yawing simulation calculation of 8 time periods, simulates the vehicle dynamic response in each time period, including vehicle acceleration, wheel slip rate, yaw rate and side slip angle, the simulation results are recorded at an interval of 10ms, forming detailed response time series data, the system analyzes the shimmy phenomenon in the braking process, calculates the vehicle vibration caused by brake pressure fluctuation, the vibration frequency mainly concentrates in 8-15Hz, the amplitude range is 0.05-0.2g, at the same time analyzes the yawing response characteristics, calculates the overshoot, rise time and settling time of the yaw rate, the typical values are 15-25%, 0.3-0.5s and 0.8-1.2 seconds, the system synthesizes the body vibration and yaw response characteristics, constructs the yawing and pitching evaluation index, the index contains two parts of time domain and frequency domain, the time domain index includes root mean square value, peak factor and waveform factor, the frequency domain index includes power spectral density and frequency band energy distribution, the system generates yawing and pitching data, the data format is three-dimensional tensor, the dimension is [sample number x time period number x characteristic number], that is [8745 x 8 x 24], wherein the characteristics include 12 time domain characteristics and 12 frequency domain characteristics, the data resolution reaches 10 ms in time dimension, 0.01 g and 0.01 degree / s in amplitude dimension, and 0.1 Hz in frequency dimension.

[0110] The multivariate regression analysis of the jitter yaw data is implemented by using an ensemble learning technique. First, the jitter yaw data is preprocessed, including missing value imputation, outlier detection and feature standardization. The K-Nearest Neighbors (KNN) imputation algorithm is used for missing value imputation, with a K value of 5 and Mahalanobis distance as the distance metric. The Local Outlier Factor (LOF) algorithm is used for outlier detection, with a neighborhood size of 20 and a threshold of 1.5. The Min-Max normalization method is used for standardization, mapping the feature values to the [0, 1] interval. The system then performs feature selection on the preprocessed data using the Recursive Feature Elimination (RFE) algorithm. The wrapper model uses a Support Vector Regression (SVR) with a Radial Basis Function (RBF) kernel, a parameter C of 10, and a parameter γ of 0.1. The feature selection process uses 5-fold cross-validation, and 12 features with the highest explanatory power are retained, including the root mean square of yaw rate, the peak value of yaw angular acceleration, and the energy proportion of the 8-12 Hz frequency band in the yaw rate power spectrum. The system then builds an ensemble of multivariate regression analysis models, including five basic regression algorithms: Ridge Regression, Lasso Regression, Elastic Net, Random Forest Regression, and Gradient Boosting Tree Regression. The Ridge Regression parameters are set to a regularization coefficient α of 0.5, the Lasso Regression parameters are set to a regularization coefficient α of 0.01, the Elastic Net parameters are set to α of 0.5 and an L1 ratio of 0.5, the Random Forest Regression parameters are set to a tree number of 100, a maximum depth of 12, and a minimum leaf node sample size of 10, and the Gradient Boosting Tree parameters are set to a tree number of 150, a learning rate of 0.05, a maximum depth of 8, and a subsampling rate of 0.8. The system trains each basic model using 70% of the total data for training and 30% for testing. The training uses 5-fold cross-validation, and the evaluation metrics include Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Determination Coefficient (R²). The performance of each model on the test set is as follows: Ridge Regression (MAE = 2.85, RMSE = 3.42, R² = 0.82), Lasso Regression (MAE = 2.92, RMSE = 3.56, R² = 0.81), Elastic Net (MAE = 2.88, RMSE = 3.48, R² = 0.82), Random Forest Regression (MAE = 2.15, RMSE = 2.75, R² = 0.88), and Gradient Boosting Tree (MAE = 1.98, RMSE = 2.58, R² = 0.89). The system integrates the prediction results of each basic model using a stacking ensemble method, with a linear SVR as the secondary learner and parameters C = 1.0 and ε = 0.1. The ensemble weights of each basic model are determined through grid search optimization, with the final weights being Ridge Regression 0.15, Lasso Regression 0.1, Elastic Net 0.15, Random Forest Regression 0.25, and Gradient Boosting Tree Regression 0.35. The final performance of the ensemble model on the test set is MAE = 1.85, RMSE = 2.42, and R² = 0.91, The system uses the trained integrated regression model to make full prediction on the jitter deviation data, generates jitter deviation regression data, the data structure is a two-dimensional matrix, the dimension is [sample number x time period number], that is, [8745 x 8], the matrix element value represents the jitter deviation severity score of each sample in each time period, the score range is 0-100, the higher the value, the more serious the jitter deviation, the system performs reliability analysis on the regression data, calculates the 95% confidence interval, the interval width is ±3.8 on average, indicating that the prediction result has high stability and reliability.

[0111] The direction deviation intervention correction angle and intervention correction vehicle speed in different time periods are progressively controlled based on the jitter deviation regression data, and the fuzzy neural network technology is used to realize it. Firstly, the jitter deviation evaluation index system is constructed, and the jitter deviation regression data is divided into 5 levels according to the severity: slight (0-20), moderate (20-40), obvious (40-60), severe (60-80) and extreme (80-100). The system establishes a fuzzy reasoning system based on the jitter deviation level, the input variables are the current jitter deviation level, the direction deviation intervention correction angle and the intervention correction vehicle speed, and the output variables are the angle progressive adjustment coefficient and the vehicle speed progressive adjustment coefficient. The input fuzzification uses triangular and trapezoidal membership functions, the jitter deviation level is divided into 5 fuzzy sets, the correction angle is divided into 7 fuzzy sets, and the correction speed is divided into 5 fuzzy sets. The output fuzzy set is the progressive adjustment coefficient, which is divided into 7 levels. The fuzzy rule base contains 175 IF-THEN rules, describing the progressive control strategy under different conditions, such as "IF the jitter deviation is severe AND the correction angle is large THEN the angle progressive adjustment coefficient is small". The reasoning method uses the Mamdani reasoning mechanism, and the defuzzification uses the centroid method. The system optimizes the basic fuzzy system through neural network, uses adaptive neuro-fuzzy inference system (ANFIS), and the network structure contains 5 layers: fuzzification layer, rule layer, normalization layer, defuzzification layer and output layer. The number of hidden layer nodes is 175, corresponding to 175 fuzzy rules. The training algorithm uses a hybrid learning method combined with least squares method and gradient descent method, the learning rate is set to 0.01, the momentum factor is 0.9, the training rounds are 500 rounds, and the error convergence threshold is 0.001. The system applies the optimized fuzzy neural network to the progressive control of 8 time periods, calculates the progressive adjustment coefficient for the direction deviation intervention correction angle and intervention correction vehicle speed in each time period, and the adjustment coefficient ranges from 0.6 to 1.2. When the jitter deviation score is high, the adjustment coefficient tends to reduce the correction, and vice versa. The system calculates the deviation progressive correction angle based on the progressive adjustment coefficient, and the calculation formula is: progressive correction angle = original correction angle × angle progressive adjustment coefficient. Similarly, the progressive correction speed is calculated, and the calculation formula is: progressive correction speed = original correction speed × vehicle speed progressive adjustment coefficient. The system sets the limit conditions for progressive correction, including angle change rate limit and vehicle speed change rate limit, the upper limit of angle change rate is 12 degrees per second, and the upper limit of vehicle speed change rate is 5 km / h per second. When the calculation result exceeds the limit, it is truncated. The system smoothes the progressive correction result using the cubic spline interpolation algorithm to ensure smooth transition between adjacent time periods. The system generates deviation progressive correction angle and progressive correction speed data, the data format is time series, the sampling interval is 10 ms, covering a 2-second control window, forming a 200-point control sequence, the data accuracy is 0.1 degrees for angle and 0.1 km / h for speed.

[0112] The braking force smooth control optimization in different time periods according to the offset progressive correction angle and the progressive correction vehicle speed is realized by using the piecewise adaptive control technology. The system first establishes the mapping relationship between the braking torque and the offset correction quantity. The piecewise linear mapping function is used. For the progressive correction angle, the mapping function is as follows: when the angle is less than 5 degrees, the braking torque increment = angle * 150 N·m / degree; when the angle is between 5-15 degrees, the braking torque increment = 750 + angle * 200 N·m / degree; when the angle is greater than 15 degrees, the braking torque increment = 2750 + angle * 50 N·m / degree. For the progressive correction vehicle speed, the mapping function is as follows: when the speed is less than 10 km / h, the braking torque increment = speed * 100 N·m / (km / h); when the speed is between 10-30 km / h, the braking torque increment = 1000 + speed * 150 N·m / (km / h); when the speed is greater than 30 km / h, the braking torque increment = 4000 + speed * 80 N·m / (km / h). The system fuses the braking torque increments of the angle mapping and the speed mapping by weighting, the weight coefficients are 0.6 and 0.4, and the basic braking torque increment is calculated. The system performs time period differentiation processing on the basic braking torque increment, constructs a time weight function, and the weight coefficients of the eight time periods are 0.3, 0.5, 0.7, 1.0, 0.9, 0.7, 0.5 and 0.3. The braking torque increment after time weighting is superimposed with the original braking torque demand data to obtain the preliminary braking torque control sequence. The system performs smooth control optimization on the braking torque control sequence, adopts a multi-stage processing strategy, applies a low-pass filter in the first stage to eliminate high-frequency fluctuations, the filter type is a Butterworth fourth-order filter, and the cutoff frequency is 5 Hz; a slope limiter is applied in the second stage to control the braking torque change rate, the rising slope limit is 600 N·m / s, and the falling slope limit is 800 N·m / s; a jitter suppressor is applied in the third stage to process small fluctuations, the jitter suppression algorithm uses a combination of dead zone control and exponential smoothing, the dead zone threshold is ±50 N·m, and the smoothing factor is 0.85, The system segments the optimized brake torque sequence after smoothing and adjusts it for different time periods to optimize the control characteristics. In the prediction segment (0-200ms), a slight pre-braking strategy is adopted, and the brake torque is limited to within 20% of the maximum demand. In the initial response segment (200-400ms), a linear growth strategy is adopted, and the brake torque growth rate is 30% / 100ms of the maximum demand. In the transition segment I (400-600ms), a rapid growth strategy is adopted, and the brake torque growth rate is 40% / 100ms of the maximum demand. In the enhancement segment (600-900ms), a stable control strategy is adopted, and the brake torque is maintained at 90-100% of the maximum demand. In the stable segment (900-1200ms), a fine-tuning control strategy is adopted, and the brake torque variation range is limited to within ±5%. In the transition segment II (1200-1500ms), a slow reduction strategy is adopted, and the brake torque reduction rate is 20% / 100ms of the maximum demand. In the weakening segment (1500-1800ms), a rapid reduction strategy is adopted, and the brake torque reduction rate is 30% / 100ms of the maximum demand. In the exit segment (1800-2000ms), a smooth exit strategy is adopted, and the brake torque is reduced to below 5% of the maximum demand. The system finally generates brake force smoothing control data, which is a multi-dimensional matrix containing brake torque time sequence, brake force distribution parameters, and control timing markers. The torque time sequence has a sampling interval of 10ms, forming a 200-point control sequence, with a torque range of 0-3500N·m and an accuracy of 1N·m. The torque distribution parameters include front and rear axle distribution ratios and left and right wheel distribution ratios. The control timing markers include boundary times for each stage and key event time points.

[0113] The application also provides a new energy vehicle safe driving control system for executing the new energy vehicle safe driving control method as described above, which comprises:

[0114] A data sampling module is configured to obtain a historical emergency avoidance data set and a corresponding user misoperation braking state through a new energy vehicle control center, perform sample undersampling processing on the historical emergency avoidance data set, and obtain an emergency avoidance balanced sample.

[0115] A brake abnormality analysis module is configured to analyze the user misoperation braking abnormality of the misoperation braking state according to the emergency avoidance balanced sample to obtain brake abnormality data, and perform brake disorder vector strength calculation in the time dimension based on the brake abnormality data to obtain disorder vector segmented clustering data.

[0116] An intervention correction perception module is configured to perform brake timing intervention correction perception according to the disorder vector segmented clustering data to generate brake intervention correction optimization data, and send the brake intervention correction optimization data to a new energy vehicle control terminal to execute new energy vehicle safe driving control.

[0117] The foregoing is considered as illustrative only of the principles of the application. Numerous modifications and changes will readily occur to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Therefore, the scope of the application is indicated by the appended claims rather than by the foregoing description, and all changes that come within the meaning and range of equivalents are intended to be embraced therein.

Claims

1. A new energy vehicle safe driving control method, characterized in that, The method comprises the following steps: Step S1: obtaining a historical emergency avoidance data set and a corresponding user misoperation braking state through a new energy vehicle control center; performing sample undersampling processing on the historical emergency avoidance data set to obtain an emergency avoidance balance sample; Step S2: performing misoperation braking abnormality analysis of the emergency avoidance scene according to the emergency avoidance balance sample to obtain braking abnormality data; Based on the braking abnormality data, the strength of the braking disordered vector in the time dimension is calculated to obtain disordered vector segmentation clustering data; Step S3: performing braking timing intervention correction perception according to the disordered vector segmentation clustering data to generate braking intervention correction optimization data; sending the braking intervention correction optimization data to a new energy vehicle control terminal to perform new energy vehicle safety driving control.

2. The new energy vehicle safe driving control method according to claim 1, characterized in that, Step S1 comprises the following steps: Step S11: obtaining a historical emergency avoidance data set and a corresponding user misoperation braking state through a new energy vehicle control center; Step S12: selecting random samples from the historical emergency avoidance data set to obtain emergency avoidance random samples; Step S13: embedding time labels into the emergency avoidance random samples to obtain emergency avoidance time label samples; Step S14: performing sample undersampling processing on the emergency avoidance time label samples to obtain an emergency avoidance balance sample.

3. The new energy vehicle safe driving control method according to claim 1, characterized in that, Step S2 comprises the following steps: Step S21: restoring the emergency avoidance scene according to the emergency avoidance balance sample; and simultaneously performing associated mapping of the misoperation braking state according to the emergency avoidance scene to obtain emergency avoidance scene-misoperation mapping data; Step S22: performing misoperation braking abnormality analysis of the emergency avoidance scene according to the emergency avoidance scene-misoperation mapping data to obtain braking abnormality data; Step S23: calculating the strength of the braking disordered vector in the time dimension based on the braking abnormality data to obtain braking disordered vector strength; Step S24: performing time segmentation clustering on the braking disordered vector strength to obtain disordered vector segmentation clustering data.

4. The new energy vehicle safe driving control method according to claim 3, characterized in that, Step S22 comprises the following steps: Extracting the throttle acceleration state and direction steering operation state of the user operation in the emergency avoidance scene-misoperation mapping data; When the acceleration of the accelerator acceleration state is greater than 12 , the duration is greater than 1.5s, and at the same time the accelerator pedal opening degree change is extracted by the engine control unit, when the accelerator pedal opening degree change is greater than 60% from the increasing rate within 1s, it is determined that the accelerator misstep abnormal acceleration state occurs. When the direction steering operation state is greater than or equal to 90°~180° within 1s, and the direction steering operation frequency is greater than or equal to 3 times of direction steering operation state within 1s, it is determined as direction misoperation abnormal steering; Based on the throttle misoperation abnormal acceleration state and the direction misoperation abnormal steering, the misoperation braking abnormality of the emergency avoidance scene is analyzed to obtain braking abnormality data.

5. The new energy vehicle safe driving control method according to claim 3, characterized in that, Step S23 comprises the following steps: Step S231: extracting the throttle misoperation abnormal acceleration change and the direction misoperation abnormal steering change angle in the time dimension of the braking abnormality data; Step S232: calculating the acceleration ratio of the throttle misoperation abnormal acceleration change, and then calculating the acceleration ratio difference within 2s; Step S233: analyzing the steering angle mutation swing of the direction misoperation abnormal steering change angle within 2s; Step S234: performing yaw impact vector strength analysis according to the acceleration ratio difference and the steering angle mutation swing to obtain the yaw impact vector strength; Step S235: nonlinear regression analysis is performed on the yaw impact vector intensity to obtain a yaw impact vector regression intensity; and a brake disorder vector intensity is calculated based on the yaw impact vector regression intensity in the time dimension to obtain the brake disorder vector intensity.

6. The new energy vehicle safe driving control method according to claim 4, characterized in that, Step S234 includes the following steps: According to the steering angle mutation swing, the yaw trajectory direction is analyzed; and the acceleration speed proportional difference is nonlinearly time-integrated to obtain acceleration speed integration data; The steering angle mutation swing is calculated to obtain a dynamic offset angle of left-right back-and-forth quick hitting directions; According to the acceleration speed integration data and the dynamic offset angle, the rotational inertia fluctuation is estimated, and then the angular momentum partial derivative is calculated to obtain rotational angular momentum partial derivative data; Based on the acceleration speed integration data, the dynamic offset angle, and the rotational angular momentum partial derivative data, the yaw trajectory direction is analyzed to obtain oblique combined vector intensity coupling data; wherein the oblique combined vector is used to describe the motion or force of the vehicle in the longitudinal and lateral directions, and the oblique combined vector contains the intensity information of the two directions; According to the oblique combined vector intensity coupling data, the yaw impact vector intensity is analyzed to obtain the yaw impact vector intensity.

7. The new energy vehicle safe driving control method according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: the disordered vector segmented clustering data is convoluted to obtain disordered vector segmented convolution data; Step S32: brake timing intervention correction sensing is performed according to the disordered vector segmented convolution data to obtain brake intervention correction data; Step S33: the brake intervention correction data is iteratively learned to generate brake intervention correction optimization data; Step S34: the brake intervention correction optimization data is sent to a new energy vehicle control terminal to perform new energy vehicle safe driving control.

8. The new energy vehicle safe driving control method according to claim 7, characterized in that, Step S32 includes the following steps: Step S321: the segmented direction offset angle and the vehicle speed stall value of different time periods in the disordered vector segmented convolution data are extracted; Step S322: the vehicle body offset angle is quantified according to the segmented direction offset angle and the vehicle speed stall value of different time periods to obtain the vehicle body offset angle; and brake torque demand data is obtained based on the vehicle body offset angle; Step S323: the brake torque demand data is controlled and optimized in different time periods based on the segmented direction offset angle and the vehicle speed stall value of different time periods to obtain brake force smoothing control data; Step S324: brake timing intervention correction sensing is performed according to the brake force smoothing control data to obtain brake intervention correction data.

9. The new energy vehicle safe driving control method according to claim 8, characterized in that, Step S323 includes the following steps: The direction offset intervention correction angle and the intervention correction vehicle speed in different time periods are determined based on the segmented direction offset angle and the vehicle speed stall value of different time periods; The vehicle intervention correction brake jittering and yawing simulation calculation is performed on the brake torque demand data according to the direction offset intervention correction angle and the intervention correction vehicle speed in different time periods to obtain jittering and yawing data; The jittering and yawing data is subjected to multiple regression analysis to generate jittering and yawing regression data; The direction deviation intervention correction angle and the intervention correction vehicle speed of different time periods are gradually controlled based on the jitter deviation regression data, to obtain a deviation gradual correction angle and a gradual correction vehicle speed; The braking force smooth control data is obtained by performing braking force smooth control optimization of different time periods according to the deviation gradual correction angle and the gradual correction vehicle speed.

10. A new energy vehicle safe driving control system, characterized in that, The new energy vehicle safety driving control system comprises: A data sampling module is configured to acquire a historical emergency avoidance data set and a corresponding user misoperation braking state through a new energy vehicle control center, and to obtain an emergency avoidance balance sample by performing sample undersampling processing on the historical emergency avoidance data set; The braking abnormality analysis module is configured to analyze the user misoperation braking abnormality of the misoperation braking state according to the emergency avoidance balance sample, to obtain braking abnormality data, and to obtain disordered vector segmentation clustering data by performing braking disordered vector strength calculation in the time dimension based on the braking abnormality data; The intervention correction perception module is configured to perform braking timing intervention correction perception according to the disordered vector segmentation clustering data, to generate braking intervention correction optimization data, and to send the braking intervention correction optimization data to a new energy vehicle control terminal, so as to execute new energy vehicle safety driving control.

Images