Automobile handling stability evaluation method based on subjective-physiological state prediction

Abstract

The invention relates to an automobile handling stability evaluation method based on subjective-physiological state prediction, and the method comprises the steps: collecting vehicle dynamic data, physiological data of a subjective evaluator and a subjective evaluation table, combining data processing and a machine learning algorithm, and predicting the handling stability of a vehicle under different driving situations. According to the method, the psychological state and subjective opinions of the driver under different driving situations can be accurately reflected through a prediction algorithm under the condition that no real person drives, and a scientific basis is provided for optimization design of vehicle control stability. The method is not only suitable for traditional fuel vehicles, but also can be widely applied to electric vehicles, hybrid electric vehicles and automatic driving vehicles, and has wide application prospects.

Description

Technical Field

[0001] The invention relates to the field of automobile chassis stability evaluation, and in particular to an automobile handling stability evaluation method based on the prediction of an evaluator's subjective-physiological state. Background Art

[0002] Vehicle handling stability (hereinafter referred to as "handling stability") is a key performance indicator to ensure that the vehicle can operate stably according to the subjective evaluator's intention under various driving conditions, and is crucial to improving the active safety of the vehicle. With the advancement of modern automobile technology and the continuous increase in speed, handling stability not only affects the driving experience, but is also directly related to driving safety. Traditional handling stability evaluation relies on actual vehicle testing, and professional evaluators are required to judge the performance of the vehicle through subjective feelings. Although this method is intuitive and effective, it is time-consuming, costly, and difficult to standardize. Especially in the rapid iteration of new vehicle model development, traditional methods are difficult to meet the needs of efficient calibration.

[0003] In order to meet the above challenges, virtual calibration technology came into being. This technology uses computer simulation to simulate real driving scenarios and predicts the dynamic behavior of the vehicle through mathematical models, thereby achieving early optimization and verification of handling stability. However, in order to make the virtual calibration results sufficiently credible, it is necessary to establish a virtual evaluation system that can accurately simulate the subjective evaluator's real vehicle evaluation. This system must not only consider objective physical parameters, but also incorporate factors of human perceptual psychology to ensure the comprehensiveness and accuracy of the simulation.

[0004] The present invention aims to propose an innovative evaluation method to address the shortcomings of existing virtual calibration technology, aiming to provide a solution that is both in line with engineering practice and close to user experience. It can accurately reflect the psychological state and subjective opinions of subjective evaluators in different driving scenarios through prediction algorithms without real people driving, thereby guiding the optimal design of vehicle handling stability and providing support for the design and development of a new generation of automobiles. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method for evaluating vehicle handling stability based on subjective-physiological state prediction, comprising the following steps:

[0006] Step 1: Design test conditions

[0007] Four test conditions are designed, including slalom, single lane change, compact double lane change, and continuous drifting conditions; covering common and uncommon dynamic operation conditions in driving, including continuous turning, emergency avoidance, rapid lane change and extreme control. Through the combination of the above conditions, the vehicle's performance in multiple dimensions such as understeer, oversteer, and yaw rate change is systematically analyzed.

[0008] Step 2: Build a data collection system

[0009] Build a data acquisition system to collect and process vehicle chassis data, biological signal data, and navigation data, and ensure accurate data transmission and time synchronization through multiple communication methods.

[0010] Specifically, the vehicle chassis signal is connected to SpeedGoat via the CAN bus to collect and transmit various dynamic data of the vehicle;

[0011] One end of SpeedGoat communicates with the PC host computer via Ethernet, and the other end is connected to an analog generator for time synchronization with the BioPac device;

[0012] The BioPac device also communicates with the PC host computer via Ethernet and is responsible for collecting biosignals such as electrocardiogram (ECG), electromyography (EMG) and electrodermal conductance (EDA). These sensors are connected to the BioPac device via wireless communication;

[0013] The DTU uses RTK signals to communicate with the remote control device, and then the remote control device is connected to the PC host computer through RS422 serial communication to provide high-precision positioning data;

[0014] The entire system integrates vehicle chassis data, biological signal data, and navigation data into the host computer through multiple communication methods (CAN bus, Ethernet, wireless communication, RS422), and finally performs data analysis and processing through MATLAB, providing a solid foundation for subsequent analysis and research.

[0015] Step 3: Data Collection

[0016] Arrange the handling stability test field according to the designed test conditions, collect data in the handling stability test field to ensure that the test conditions can truly reflect the actual driving environment; collect vehicle dynamic data, high-precision positioning data and physiological state data of the subjective evaluator. After each test, the subjective evaluator needs to fill in the subjective evaluation form.

[0017] Specifically, the test scenarios in the designed test conditions are arranged, including but not limited to straight roads, curves, sharp turns, and roads with different friction coefficients, so as to simulate a variety of driving conditions, so as to comprehensively evaluate the vehicle's handling stability and performance; the subjective evaluator drives a specially modified test vehicle equipped with an advanced data acquisition system according to the established test plan, completes a series of predetermined test tasks, and records various dynamic data of the vehicle in real time; at the same time, high-precision positioning data is recorded through the remote control equipment; in addition, the biological signal data of the subjective evaluator's physiological state is also synchronously recorded through the BioPac system; after each test condition is completed, in addition to recording the objective data, the subjective evaluator is also required to fill out a detailed subjective evaluation form, which includes scoring and evaluation of steering, yaw, controllability, roll, lateral direction, and speed, aiming to capture the subjective evaluator's immediate feelings about the vehicle performance.

[0018] Step 4: Data processing

[0019] The collected data is preprocessed, including time synchronization, data frequency reduction, smoothing and outlier removal; then the characteristic value calculation is performed, including vehicle dynamic characteristics and physiological signal characteristics.

[0020] Specifically, the first step is data preprocessing:

[0021] The first step is to synchronize the time to ensure that the data recorded by all sensors and equipment have a unified time base. In this step, the analog receiving ports of Biopac and the vehicle data acquisition system are connected to the same analog generator, and an analog signal is sent before each test. The rising edge of this signal is the first point where the test starts.

[0022] The second step is to reduce the frequency of the data and halve the frequency of the physiological data;

[0023] The third step is to smooth the down-converted data to eliminate random fluctuations and small changes in the data, making the signal more stable and continuous. The moving average method is used for smoothing. The formula of the moving average method is:

[0024]

[0025] Where y t ,y t-1 , ...represent the observation values ​​of t, t-1, ... respectively; N is the number of average items;

[0026] Step 4: Check whether there are outliers in the data. Outliers may be caused by equipment failure, environmental interference or other abnormal factors. After smoothing, these outliers are identified and eliminated by the standard deviation method to ensure the purity of the data. Data that is more than twice the standard deviation from the mean is considered an outlier and is eliminated. The calculation formula for the standard deviation is:

[0027]

[0028] Among them, σ represents the standard deviation, x i represents the i-th data point, μ represents the mean of the data; N represents the total number of data points;

[0029] The data processed through the above steps finally forms the preprocessing result.

[0030] After obtaining the preprocessing results, characteristic values ​​are calculated according to the preprocessing results, including vehicle dynamic characteristics and physiological signal characteristics.

[0031] The dimensions of vehicle dynamic characteristics include: one or more of longitudinal, lateral, yaw, roll, pitch, steering wheel, and steering; the indicators of the longitudinal dimension include initial speed, speed decay, and average deceleration; the indicators of the lateral dimension include maximum lateral acceleration and maximum lateral jerk; the indicators of the yaw dimension include maximum yaw angle and average yaw angular velocity; the indicators of the roll dimension include maximum roll angle and average roll change rate; the indicators of the pitch dimension include maximum pitch angle; the indicators of the steering wheel dimension include average steering wheel adjustment amount and average steering wheel adjustment speed; the indicators of the steering dimension include understeering factor and oversteering factor;

[0032] Indicators of physiological signal characteristics include: average heart rate and SCL growth rate;

[0033] The indicators of subjective evaluation characteristics include: one or more of steering evaluation, yaw evaluation, controllability evaluation, roll evaluation, lateral evaluation, and speed evaluation;

[0034] The calculation process of the understeering factor index in the vehicle dynamic characteristics is as follows:

[0035] (1) The formula for calculating the center of mass sideslip angle is:

[0036]

[0037] Where β is the sideslip angle of the center of mass, Vy is the lateral velocity, and Vx is the longitudinal velocity;

[0038] (2) The calculation formula of the front wheel slip angle is:

[0039]

[0040] In the formula, αf is the front wheel slip angle, w r is the yaw angular velocity, l f is the distance from the center of mass to the front axle, δ f is the front wheel turning angle;

[0041] (3) The formula for calculating the rear wheel slip angle is:

[0042]

[0043] In the formula, α r is the front wheel slip angle, w r is the yaw angular velocity, l r is the distance from the center of mass to the rear axle;

[0044] The remaining vehicle characteristic values ​​are the mean or maximum values ​​within the extracted operating conditions;

[0045] After completing the processing of the vehicle dynamic feature data, the NeuroKit2 library in the Python environment is used to process the electrocardiogram, and NeuroKit2 is used to denoise the ECG signal, and then the instantaneous heart rate is calculated; then the skin electrode activity signal is processed, first using NeuroKit2 for denoising, and then the skin electrode level is extracted.

[0046] After completing the extraction of vehicle dynamic feature data and physiological signal features, the subjective evaluation data of the subjective evaluator is recorded.

[0047] Step 5: Correlation Analysis

[0048] The correlation analysis method is used to analyze the correlation between vehicle-related features, physiological signal features and subjective evaluation data, and significantly correlated features are screened out as input to the prediction model.

[0049] Specifically, the correlation analysis was performed using the Pearson correlation analysis method in the Origin software, and the steps are as follows:

[0050] First, calculate the correlation coefficient r:

[0051]

[0052] Among them, x i and i represent the observed values ​​of two variables, represents the mean of two variables;

[0053] The strength of the linear relationship between the two variables is evaluated based on the calculated correlation coefficient r. If the correlation is closer to 1, the positive correlation is higher, and if the correlation is closer to -1, the negative correlation is higher. The range of r values ​​indicating the correlation is as follows:

[0054]

[0055] Secondly, the calculated r value needs to be tested for statistical significance to determine whether the observed correlation is not due to random fluctuations. The t-test method is used, and its statistic is:

[0056]

[0057] Where n is the sample size;

[0058] Find the corresponding critical value p according to the degrees of freedom df=n-2; if the p value is less than the set significance level, the correlation is considered to be statistically significant. If the p value is greater than the set significance level, it means that the existing data is insufficient to prove that there is a significant linear relationship between the two variables; then eliminate the parameters that fail the significance test based on the results.

[0059] Step 6: Training and evaluation of prediction models

[0060] According to the correlation analysis results, vehicle-related features with significant linear relationships between physiological and subjective evaluation data are used as inputs of the prediction model for training; the prediction model uses the random forest method in machine learning methods, and verifies the accuracy of the model by calculating the prediction errors of the training set and the test set.

[0061] The specific steps are as follows:

[0062] The first step is to standardize or normalize all input features to make data of different dimensions comparable. The normalization formula is:

[0063]

[0064] Where X is the original data value, X min and X max is the minimum and maximum value of the feature in the data set, X norm is the normalized value;

[0065] The second step is to divide the data set into training set and test set;

[0066] The third step is to create a random forest model. The steps are as follows:

[0067] First, Bootstrap sampling is performed to extract samples with replacement from the original data to create a training set. Assuming the size of the data set is N, the probability of each sampling is 1 / N;

[0068] Secondly, at each split, a portion of all features is randomly selected for evaluation;

[0069] Then a single decision tree is constructed. The splitting criterion of the decision tree is based on the mean square error between the training sample and the predicted value. The formula is:

[0070]

[0071] Among them, y i is the true value, is the predicted value, n is the number of samples;

[0072] When the maximum depth of the decision tree is reached, the node is marked as a leaf node and assigned an average value;

[0073] Finally, the mean of all decision tree results is calculated using the formula:

[0074]

[0075] Where T is the number of trees, and h(x) represents the prediction result of the tth tree for the sample;

[0076] The fourth step is to evaluate the model and calculate the prediction error of the model's training set and test set;

[0077] Step 7: Subjective Evaluation Prediction

[0078] Based on the physiological signal characteristics, the mental workload scores of subjective evaluators are predicted, and the predicted results are compared with the actual test results to verify the effectiveness.

[0079] Beneficial effects of the present invention:

[0080] The present invention aims at the deficiencies in the existing virtual calibration technology. By combining the physiological data and subjective evaluation table of the subjective evaluator, a method for evaluating the handling stability of an automobile based on the prediction of the subjective-physiological state is proposed. The method combines data processing and machine learning algorithms to predict the handling stability of the vehicle in different driving situations, which is both in line with engineering practice and close to user experience, and provides strong support for the design and development of a new generation of automobiles. The method can accurately reflect the psychological state and subjective opinions of the subjective evaluator in different driving situations through the prediction algorithm without relying on real-life driving, and provide a scientific basis for the optimization design of vehicle handling stability. The present invention not only has important applications in the design and development of traditional fuel vehicles, but can also be widely used in various types of motor vehicles, including but not limited to electric vehicles, hybrid vehicles and self-driving vehicles, and has broad application prospects. At the same time, the method proposed by the present invention can also be used for the evaluation of other vehicle performance based on the scoring of subjective evaluators, such as braking performance, power performance, smoothness, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is a schematic diagram of the overall process of the evaluation method of the present invention;

[0082] Figure 2 It is a schematic diagram of a serpentine working condition in a specific embodiment of the present invention;

[0083] Figure 3 It is a schematic diagram of a single lane shifting condition in a specific embodiment of the present invention;

[0084] Figure 4 It is a schematic diagram of a compact double lane shifting condition in a specific implementation manner of the present invention;

[0085] Figure 5 A schematic diagram of building a data acquisition system of the present invention;

[0086] Figure 6 It is a schematic diagram of the data processing flow of the present invention;

[0087] Figure 7 This is a schematic diagram of the correlation analysis results in a specific embodiment of the present invention;

[0088] Figure 8 This is a schematic diagram of the overall process of the random forest model of the present invention;

[0089] Fig. 9 It is a radar chart of predicted evaluation scores and actual evaluation scores in a specific implementation manner of the present invention. DETAILED DESCRIPTION

[0090] like Figure 1 As shown, the present invention provides a method for evaluating vehicle handling stability based on subjective-physiological state prediction, comprising the following steps:

[0091] Step 1: Design test conditions

[0092] With reference to GBT 6323-2014 and ISO 3888-2, four test conditions are designed, including slalom, single lane change, compact double lane change, and continuous drifting. These four conditions cover common and uncommon dynamic operation situations in driving, including continuous turning (slalom), emergency avoidance (single lane change), rapid lane change (compact double lane change) and extreme control (continuous drifting).

[0093] Specifically, the path design of the serpentine slalom test is as follows: Figure 2 As shown in the figure, L is the distance between the two poles, which is set to 18 meters. The tests are carried out on high-adhesion roads and low-adhesion roads respectively. The test vehicle speed starts from a sufficiently low speed to ensure the stability of the vehicle when changing lanes, and then gradually increases the speed around the poles until the limit is reached.

[0094] The specific route design of the single lane shift test is as follows: Figure 3As shown, there are two test conditions: high adhesion to low adhesion road surface and low adhesion to high adhesion road surface. The test vehicle speed starts from a sufficiently low speed to ensure the stability of the vehicle when changing lanes, and then gradually increases the lane changing speed until the limit is reached.

[0095] The specific route design of the compact double lane shift test is as follows: Figure 4 As shown, the test vehicle slowly increases the test vehicle speed step by step until the vehicle becomes unstable.

[0096] The continuous drift test is conducted in a dedicated vehicle handling stability test field, covering two different ground condition changes: high adhesion to low adhesion and low adhesion to high adhesion. During the whole process, ESC is always kept off, the vehicle enters the test site at a speed of 90-100 km / h, and achieves steady-state drift through throttle control while keeping the vehicle moving within the test area, simulating the psychological and physiological reactions of the subjective evaluator when the vehicle is in an extreme state.

[0097] Through the combination of the above working conditions, the vehicle's performance in multiple dimensions such as understeer, oversteer, and yaw rate change can be systematically analyzed.

[0098] Step 2: Build a data collection system

[0099] like Figure 5 As shown, a data acquisition system is built to efficiently collect and process vehicle chassis data, biological signal data, and navigation data, and ensure accurate data transmission and time synchronization through multiple communication methods.

[0100] Specifically, the vehicle chassis signal is connected to SpeedGoat via the CAN bus to collect and transmit various dynamic data of the vehicle;

[0101] One end of SpeedGoat communicates with the PC host computer via Ethernet, and the other end is connected to an analog generator for time synchronization with the BioPac device;

[0102] The BioPac device also communicates with the PC host computer via Ethernet and is responsible for collecting biosignals such as electrocardiogram (ECG), electromyography (EMG) and electrodermal conductance (EDA). These sensors are connected to the BioPac device via wireless communication;

[0103] The DTU uses RTK signals to communicate with the remote control device, and then the remote control device is connected to the PC host computer through RS422 serial communication to provide high-precision positioning data;

[0104] The entire system integrates vehicle chassis data, biological signal data, and navigation data into the host computer through multiple communication methods (CAN bus, Ethernet, wireless communication, RS422), and finally performs data analysis and processing through MATLAB, providing a solid foundation for subsequent analysis and research.

[0105] Step 3: Data Collection

[0106] The team arranges the handling stability test field in strict accordance with the designed test conditions, collects data at the handling stability test field, and ensures that the test conditions can truly reflect the actual driving environment; collects vehicle dynamic data, high-precision positioning data, and physiological state data of subjective evaluators. After each test, the subjective evaluator needs to fill out a subjective evaluation form.

[0107] Specifically, tools such as pile barrels and tape are used to restore the test scenarios in the designed test conditions, including but not limited to straight roads, curves, sharp turns, and roads with different friction coefficients, etc., to simulate a variety of driving conditions, so as to comprehensively evaluate the vehicle's handling stability and performance; the subjective evaluator drives a specially modified test vehicle equipped with an advanced data acquisition system according to the established test plan, completes a series of predetermined test tasks, and records various dynamic data of the vehicle in real time; at the same time, high-precision positioning data is recorded through the remote control equipment; in addition, the biological signal data of the subjective evaluator's physiological state is also synchronously recorded through the BioPac system; after each test condition is completed, in addition to recording objective data, the subjective evaluator is also required to fill out a detailed subjective evaluation form, as shown in Table 1. This form is designed to capture the subjective evaluator's immediate feelings about the vehicle's performance, covering steering, controllability, yaw control and other aspects of the vehicle's operating characteristics.

[0108] Table 1 Subjective evaluation table

[0109] 1 point 2 points 3 points 4 points 5 points Turn to evaluation Swing Evaluation Controllability evaluation Roll evaluation Lateral evaluation Speed ​​evaluation

[0110] Step 4: Data processing

[0111] The collected data is preprocessed, including time synchronization, data frequency reduction, smoothing and outlier removal; then the characteristic value is calculated, including vehicle dynamic characteristics and physiological signal characteristics. Figure 6 The data is processed according to the flow shown.

[0112] The first step is data preprocessing:

[0113] The first step is to synchronize the time to ensure that the data recorded by all sensors and equipment have a unified time base. In this step, the analog receiving ports of Biopac and the vehicle data acquisition system are connected to the same analog generator, and an analog signal is sent before each test. The rising edge of this signal is the first point where the test starts.

[0114] The second step is to reduce the frequency of the data. Since the acquisition frequency of the Biopac device is 2000Hz, and the acquisition frequency of the vehicle data acquisition system is only 1000Hz, the frequency of the physiological data is halved;

[0115] The third step is to smooth the down-converted data. This process aims to eliminate random fluctuations and small changes in the data to make the signal more stable and continuous. The moving average method is used for smoothing. The formula of the moving average method is as follows:

[0116]

[0117] Where y t ,y t-1 ,...represent the observation values ​​of t, t-1,... respectively; N is the number of average items.

[0118] Step 4: Check whether there are outliers in the data. Outliers may be caused by equipment failure, environmental interference or other abnormal factors. After smoothing, these outliers are identified and eliminated by the standard deviation method to ensure the purity of the data. Data that is more than twice the standard deviation from the mean is considered an outlier and is eliminated. The standard deviation calculation formula is as follows:

[0119]

[0120] Among them, σ represents the standard deviation, x i represents the i-th data point, μ represents the mean of the data; N represents the total number of data points.

[0121] The data processed through the above steps finally forms the preprocessing result.

[0122] After obtaining the preprocessing results, characteristic values ​​are calculated based on the preprocessing results, including vehicle dynamic characteristics and physiological signal characteristics.

[0123] The selection of vehicle dynamic features is shown in Table 2, and the selection of physiological signal features is shown in the first two items in Table 3:

[0124] Table 2 Characteristics of vehicle signals

[0125]

[0126] Table 3 Physiological and subjective evaluation characteristics

[0127] Dimensions Indicator 1 physiological Average heart rate physiological SCL Growth Rate subjective Turn to evaluation subjective Swing Evaluation subjective Controllability evaluation subjective Roll evaluation subjective Lateral evaluation subjective Speed ​​evaluation

[0128] Since the handling stability test field is usually designed as a flat field without slopes, the present embodiment does not consider vertical movement (i.e. up and down bumps), which can simplify the test conditions and focus on the dynamic characteristics of other degrees of freedom; and in the working conditions involved, no acceleration or braking operations are involved, so the pitch angle change of the vehicle will be very small, and the measurement results may be affected by large noise, so the change in the pitch angle is ignored to reduce the uncertainty in data processing.

[0129] The purpose of selecting physiological signals is: electrocardiogram mainly reflects changes in heart activity, while electrodermal signal can be used as an indicator of the activity of the sympathetic nervous system. Combining these two physiological signals can capture the body's response to psychological load from different angles.

[0130] The calculation process of understeering factor in vehicle dynamic characteristics is as follows:

[0131] (1) The formula for calculating the center of mass sideslip angle is:

[0132]

[0133] Where β is the sideslip angle of the center of mass, Vy is the lateral velocity, and Vx is the longitudinal velocity;

[0134] (2) The calculation formula of the front wheel slip angle is:

[0135]

[0136] In the formula, α f is the front wheel slip angle, w r is the yaw angular velocity, l f is the distance from the center of mass to the front axle, δ f is the front wheel turning angle;

[0137] (3) The formula for calculating the rear wheel slip angle is:

[0138]

[0139] In the formula, α r is the front wheel slip angle, w r is the yaw angular velocity, l r is the distance from the center of mass to the rear axle.

[0140] Since the remaining vehicle characteristic values ​​are all average values ​​or maximum values ​​within the extracted operating conditions, the calculation method is prior art and will not be described here in detail.

[0141] After completing the processing of the vehicle dynamic feature data, the NeuroKit2 library in the Python environment is used to process the electrocardiogram, and NeuroKit2 is used to denoise the ECG signal, and then the instantaneous heart rate is calculated; then the skin electrode activity signal is processed, first using NeuroKit2 for denoising, and then the skin electrode level is extracted.

[0142] After completing the extraction of vehicle dynamic feature data and physiological signal features, the physiological and subjective evaluation feature table in Table 3 is used to record the subjective evaluation data of the subjective evaluator.

[0143] Step 5: Correlation Analysis

[0144] The Pearson correlation analysis method is used to analyze the correlation between vehicle-related features, physiological signal features and subjective evaluation data, and significantly correlated features are screened out as input to the prediction model.

[0145] Specifically, the correlation analysis was performed using the Pearson correlation analysis method in the Origin software, and the steps are as follows:

[0146] First, calculate the correlation coefficient r:

[0147]

[0148] Among them, x i and i represent the observed values ​​of two variables, represents the mean of two variables;

[0149] According to the calculated correlation coefficient r, the strength of the linear relationship between the two variables is evaluated, such as Figure 7 As shown, if the correlation is closer to 1, the positive correlation is higher, and if the correlation is closer to -1, the negative correlation is higher; the r value indicates that the value range of the correlation is as follows:

[0150]

[0151] Secondly, the calculated r value needs to be tested for statistical significance to determine whether the observed correlation is not due to random fluctuations. The present invention adopts the t-test method, and its statistic is:

[0152]

[0153] Where n is the sample size;

[0154] According to the degree of freedom df=n-2, the corresponding critical value p is found; if the p value is less than the set significance level (0.01 is used in this embodiment), the correlation is considered to be statistically significant; if the p value is greater than the set significance level, it means that the existing data is insufficient to prove that there is a significant linear relationship between the two variables; then the parameters that failed the significance test are eliminated according to the results.

[0155] Step 6: Training and evaluation of prediction models

[0156] According to the correlation analysis results, the vehicle-related features that have a significant linear relationship with the physiological or subjective evaluation data in Table 3 are used as the input of the prediction model for training; the prediction model uses the random forest method in the machine learning method, and the random forest model process is as follows Figure 8 As shown; by calculating the prediction errors of the training set and the test set, the accuracy of the model is verified. The specific steps are as follows:

[0157] The first step is to standardize or normalize all input features to make data of different dimensions comparable. The normalization formula is:

[0158]

[0159] Where X is the original data value, X min and X max is the minimum and maximum value of the feature in the data set, X norm is the normalized value;

[0160] The second step is to divide the data set into a training set and a test set. In this embodiment, 90% is used for training and 10% is used for testing.

[0161] The third step is to create a random forest model, which is implemented as follows:

[0162] First, Bootstrap sampling is performed to extract samples with replacement from the original data to create a training set. Assuming the size of the data set is N, the probability of each sampling is 1 / N;

[0163] Secondly, at each split, a portion of all features is randomly selected for evaluation. In this embodiment, the number of selected features is set to 3;

[0164] Then a single decision tree is constructed. The splitting criterion of the decision tree is based on the mean square error between the training sample and the predicted value. The formula is:

[0165]

[0166] Among them, y i is the true value, is the predicted value, n is the number of samples;

[0167] When the maximum depth of the decision tree is reached, the node is marked as a leaf node and assigned an average value;

[0168] Finally, the mean of all decision tree results is calculated using the formula:

[0169]

[0170] Where T is the number of trees, and h(x) represents the prediction result of the tth tree for the sample.

[0171] The fourth step is to evaluate the model and calculate the prediction error of the model's training set and test set. The results are shown in Table 4.

[0172] Table 4 Random forest prediction error

[0173] Evaluation characteristics Test set root mean square error Root mean square error of training set Average heart rate 3.4339 2.8109 SCL Growth Rate 0.62958 0.34394 Turn to evaluation 0.19931 0.18966 Swing Evaluation 0.4601 0.18224 Controllability evaluation 0.11404 0.17878 Roll evaluation 0.25017 0.20859 Lateral evaluation 0.28869 0.23895 Speed ​​evaluation 0.32516 0.23935

[0174] Step 7: Subjective Evaluation Prediction

[0175] According to the physiological signal characteristics, the mental workload scores of the subjective evaluators are predicted, and the predicted results are compared with the actual test results to verify the effectiveness of the present invention.

[0176] The physiological signals of the resting state, stable driving state, and intense driving state in the test conditions were selected as 5 points, 3 points, and 1 point for the mental workload score, respectively. At the same time, the binary points between the three states were selected as 4 points and 2 points, so as to derive the mental workload score from the physiological signal characteristics. The predicted subjective evaluation results and the subjective evaluation results obtained from the actual test were visualized as shown in the figure. Fig. 9 shown.

Claims

1. A method for evaluating vehicle handling stability based on subjective-physiological state prediction, characterized in that: The steps include: Step 1: Design test conditions Design test conditions, including one or more of the following conditions: serpentine slalom, single lane change, compact double lane change, and continuous drifting; cover dynamic operation conditions during driving, including one or more of the following conditions: continuous turning, emergency avoidance, rapid lane change, and extreme control; Step 2: Build a data collection system Build a data acquisition system to collect and process vehicle chassis data, biosignal data, and navigation data, and ensure accurate data transmission and time synchronization through multiple communication methods; Step 3: Data Collection Arrange the handling stability test field according to the designed test conditions, collect data in the handling stability test field to ensure that the test conditions can truly reflect the actual driving environment; collect vehicle dynamic data, high-precision positioning data and driver's physiological status data. After each test, the driver needs to fill in a subjective evaluation form; Step 4: Data processing Preprocess the collected data, including time synchronization, data frequency reduction, smoothing and outlier removal; Then the characteristic value calculation is performed, including the vehicle dynamic characteristics and physiological signal characteristics; Step 5: Correlation Analysis The Pearson correlation analysis method is used to analyze the correlation between vehicle-related features, physiological signal features, and subjective evaluation data, and significantly correlated features are selected as the input of the prediction model. The steps are as follows: First, calculate the correlation coefficient r: Among them, x i and i represent the observed values ​​of two variables, represents the mean of two variables; The strength of the linear relationship between the two variables is evaluated based on the calculated correlation coefficient r. If the correlation is closer to 1, the positive correlation is higher, and if the correlation is closer to -1, the negative correlation is higher. The range of r values ​​indicating the correlation is as follows: |r|>0.7, strong correlation 0.4<|r|≤0.7, medium correlation 0.2|r|≤0.4, weak correlation |r|≤0.2, almost no correlation Secondly, the calculated r value needs to be tested for statistical significance to determine whether the observed correlation is not due to random fluctuations. The t-test method is used, and its statistic is: Where n is the sample size; Find the corresponding critical value p according to the degree of freedom df=n-2; if the p value is less than the set significance level, the correlation is considered statistically significant; if the p value is greater than the set significance level, it means that the existing data is insufficient to prove that there is a significant linear relationship between the two variables; then eliminate the parameters that failed the significance test according to the results; Step 6: Training and evaluation of prediction models According to the correlation analysis results, the vehicle-related features with significant linear relationships between physiological and subjective evaluation data are used as inputs of the prediction model for training; the prediction model uses the random forest method in the machine learning method, and verifies the accuracy of the model by calculating the prediction errors of the training set and the test set; Step 7: Subjective Evaluation Prediction Based on the physiological signal characteristics, the mental workload scores of subjective evaluators are predicted, and the predicted results are compared with the actual test results to verify the effectiveness.

2. The method for evaluating vehicle handling stability based on subjective-physiological state prediction according to claim 1, characterized in that: In step 2, the vehicle chassis signal is connected to SpeedGoat via the CAN bus to collect and transmit various dynamic data of the vehicle; One end of SpeedGoat communicates with the PC host computer via Ethernet, and the other end is connected to an analog generator for time synchronization with the BioPac device; The BioPac device also communicates with the PC host computer via Ethernet and is responsible for collecting ECG, EMG and skin electrical signals. These sensors are connected to the BioPac device via wireless communication; The DTU uses RTK signals to communicate with the remote control device, and then the remote control device is connected to the PC host computer through RS422 serial communication to provide high-precision positioning data; The entire system integrates vehicle chassis data, biological signal data, and navigation data into the host computer, and ultimately performs data analysis and processing through MATLAB.

3. The method for evaluating vehicle handling stability based on subjective-physiological state prediction according to claim 1, characterized in that: In step three, the test scenarios in the designed test conditions are arranged, including but not limited to straight roads, curves, sharp turns and roads with different friction coefficients, so as to simulate various driving conditions; the driver drives the specially modified test vehicle equipped with an advanced data acquisition system according to the established test plan, completes a series of predetermined test tasks, and records various dynamic data of the vehicle in real time; at the same time, high-precision positioning data is recorded through the remote control equipment; the biological signal data of the driver's physiological state is synchronously recorded through the BioPac system; after each test condition is completed, in addition to recording objective data, the driver is also required to fill out a detailed subjective evaluation form, which includes scoring and evaluation of steering, yaw, controllability, roll, lateral direction, and speed, and records the driver's immediate feelings about the vehicle performance.

4. The method for evaluating vehicle handling stability based on subjective-physiological state prediction according to claim 1, characterized in that: The data preprocessing steps in step 4 are as follows: The first step is to synchronize the time to ensure that the data recorded by all sensors and equipment have a unified time base. In this step, the analog receiving ports of Biopac and the vehicle data acquisition system are connected to the same analog generator, and an analog signal is sent before each test. The rising edge of this signal is the first point where the test starts. The second step is to reduce the frequency of the data and halve the frequency of the physiological data; The third step is to smooth the down-converted data to eliminate random fluctuations and small changes in the data, making the signal more stable and continuous. The moving average method is used for smoothing. The formula of the moving average method is: Where y t ,y t-1 , ...represent the observation values ​​of t, t-1, ... respectively; N is the number of average items; Step 4: Check whether there are outliers in the data. Outliers may be caused by equipment failure, environmental interference or other abnormal factors. After smoothing, these outliers are identified and eliminated by the standard deviation method to ensure the purity of the data. Data that is more than twice the standard deviation from the mean is considered an outlier and is eliminated. The calculation formula for the standard deviation is: Among them, σ represents the standard deviation, x i represents the i-th data point, μ represents the mean of the data; N represents the total number of data points; The data processed through the above steps finally forms the preprocessing result.

5. The method for evaluating vehicle handling stability based on subjective-physiological state prediction according to claim 1, characterized in that: The steps for calculating the eigenvalues ​​in step 4 are as follows: The dimensions of vehicle dynamic characteristics include: one or more of longitudinal, lateral, yaw, roll, pitch, steering wheel, and steering; the indicators of the longitudinal dimension include initial speed, speed decay, and average deceleration; the indicators of the lateral dimension include maximum lateral acceleration and maximum lateral jerk; the indicators of the yaw dimension include maximum yaw angle and average yaw angular velocity; the indicators of the roll dimension include maximum roll angle and average roll change rate; the indicators of the pitch dimension include maximum pitch angle; the indicators of the steering wheel dimension include average steering wheel adjustment amount and average steering wheel adjustment speed; the indicators of the steering dimension include understeering factor and oversteering factor; Indicators of physiological signal characteristics include: average heart rate and SCL growth rate; The indicators of subjective evaluation characteristics include: one or more of steering evaluation, yaw evaluation, controllability evaluation, roll evaluation, lateral evaluation, and speed evaluation; The calculation process of the understeering factor index in the vehicle dynamic characteristics is as follows: (1) The formula for calculating the center of mass sideslip angle is: Where β is the sideslip angle of the center of mass, Vy is the lateral velocity, and Vx is the longitudinal velocity; (2) The calculation formula of the front wheel slip angle is: In the formula, α f is the front wheel slip angle, w r is the yaw angular velocity, l f is the distance from the center of mass to the front axle, δ f is the front wheel turning angle; (3) The calculation formula of the rear wheel slip angle is: In the formula, α r is the front wheel slip angle, w r is the yaw angular velocity, l r is the distance from the center of mass to the rear axle; The remaining vehicle characteristic values ​​are the mean or maximum values ​​within the extracted operating conditions; After the vehicle dynamic feature data is processed, the ECG signal is denoised and the instantaneous heart rate is calculated. The skin electrical activity signal is denoised and the skin electrical level is extracted. After completing the extraction of vehicle dynamic feature data and physiological signal features, the subjective evaluation data of the subjective evaluator is recorded.

6. The method for evaluating vehicle handling stability based on subjective-physiological state prediction according to claim 1, characterized in that: Step 6 The method for training and evaluating the prediction model is as follows: The first step is to standardize or normalize all input features to make data of different dimensions comparable. The normalization formula is: Where X is the original data value, X min and X max is the minimum and maximum value of the feature in the data set, X norm is the normalized value; The second step is to divide the data set into training set and test set; The third step is to create a random forest model. The steps are as follows: First, Bootstrap sampling is performed to extract samples with replacement from the original data to create a training set. Assuming the size of the data set is N, the probability of each sampling is 1 / N; Secondly, at each split, a portion of all features is randomly selected for evaluation; Then a single decision tree is constructed. The splitting criterion of the decision tree is based on the mean square error between the training sample and the predicted value. The formula is: Among them, y i is the true value, is the predicted value, n is the number of samples; When the maximum depth of the decision tree is reached, the node is marked as a leaf node and assigned an average value; Finally, the mean of all decision tree results is calculated using the formula: Where T is the number of trees, and h(x) represents the prediction result of the tth tree for the sample; The fourth step is to evaluate the model and calculate the prediction error of the model's training set and test set.

Images