Method and system for evaluating motion sickness comfort of electric passenger car

Abstract

The invention relates to the technical field of passenger car evaluation, in particular to an electric passenger car motion sickness comfort evaluation method and system. The method comprises the following steps: the same fixed driver continuously drives the tested electric passenger car to a target number of turns according to a preset cycle condition; respectively fixing the motion sickness simulation robots on a co-driver seat and a rear row right seat of the tested vehicle; for the same driver, the same tested vehicle and the same cycle condition, the test steps are repeated for multiple times, and the arithmetic mean value of data collected for multiple times is taken as a multi-dimensional physical stimulation vector; and inputting the multi-dimensional physical stimulation vector into a motion sickness prediction model which is pre-calibrated and subjected to cross validation, outputting predicted motion sickness scores of a co-driver position and a rear row right side position, and if the predicted motion sickness scores are higher than a preset comfort threshold score, judging that the motion sickness comfort of the detected electric passenger vehicle at the position is unqualified, otherwise, judging that the motion sickness comfort of the detected electric passenger vehicle is not qualified. And reversely positioning the key physical stimulation component causing the over-threshold. According to the technical scheme, the motion sickness comfort of the electric passenger car can be accurately evaluated and traced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of passenger car evaluation technology, in particular to a method and system for evaluating the motion sickness comfort of an electric passenger car. BACKGROUND

[0002] Motion sickness is a common physiological phenomenon that occurs when a passenger is in a vehicle. When the passenger's body is subjected to up-and-down jolts, forward-and-back shaking, left-and-right swaying, or acceleration and deceleration, a series of uncomfortable physiological reactions are triggered. These reactions include, but are not limited to, dizziness, cold sweat, headache, stomach discomfort, nausea, and vomiting, which severely affect the passenger's riding experience and physical health. In modern society, where transportation is increasingly frequent, motion sickness has become an important factor affecting people's choice of transportation and riding comfort.

[0003] With the rapid development of the automotive industry, pure electric passenger cars have gained increasing market share due to their environmental friendliness and energy efficiency. However, compared to traditional gasoline vehicles, the unique power system of electric vehicles poses new challenges to riding comfort. The instantaneous high-torque characteristics of electric vehicles enable rapid power output during starting and acceleration, which can easily disrupt the body's balance system and trigger motion sickness. At the same time, the energy recovery braking system produces a different braking sensation than traditional gasoline vehicles, and the energy recovery feedback during braking can also cause passengers to feel abnormal motion stimuli, increasing the probability of motion sickness. Relevant data show that the incidence of motion sickness among electric vehicle passengers is significantly higher than that among gasoline vehicle passengers.

[0004] The market for pure electric passenger cars is in a rapid development stage, with their sales share of the overall passenger car market continuously increasing. As the market size continues to expand, consumers' demands for electric vehicle riding comfort are also increasing. Motion sickness is one of the key factors affecting riding comfort.

[0005] In view of the increasingly prominent problem of motion sickness in electric vehicles, there is an urgent need for a scientific and rigorous testing and evaluation method. Currently, although there are some testing methods for vehicle riding comfort, most lack a specialized testing and evaluation system for motion sickness. Existing testing methods often fail to accurately quantify vehicle indicators strongly related to motion sickness, making it difficult to comprehensively and objectively evaluate the motion sickness comfort of electric vehicles in different seating positions. SUMMARY

[0006] The purpose of the present application is to provide a method and system for evaluating the motion sickness comfort of an electric passenger car, which can accurately evaluate the motion sickness comfort of an electric passenger car and trace the source.

[0007] In order to achieve the above-mentioned purpose, the first aspect, the present application provides a method for evaluating the motion sickness comfort of an electric passenger car, comprising: The measured electric passenger car is continuously driven by the same fixed driver under a preset cycle working condition to a target number of laps; The motion sickness simulation robot is fixed on the co-driver seat and the right rear seat of the measured car respectively; The same driver, the same measured car, and the same cycle working condition are repeatedly tested several times, and the arithmetic mean of the data collected several times is taken as the multi-dimensional physical stimulation vector of the position; The multi-dimensional physical stimulation vector is input into a pre-calibrated and cross-validated motion sickness prediction model, and the motion sickness prediction model takes the physical stimulation vector of the robot as the input and takes the motion sickness score as the output; The motion sickness prediction model directly outputs the predicted motion sickness score of the co-driver position and the right rear position, and if the predicted motion sickness score is higher than the preset comfort threshold score, it is determined that the motion sickness comfort of the measured electric passenger car at the position is unqualified, and the key physical stimulation component leading to the threshold is located in reverse.

[0008] The beneficial effects of the basic scheme: the same fixed driver and the preset cycle working condition are used to ensure that the driving behaviors (acceleration / deceleration intensity, steering smoothness, vehicle speed control accuracy, etc.) of different test batches and different positions are completely consistent, avoid the input fluctuation of physical stimulation caused by the difference in driving habits of the driver, and make the evaluation result only related to the dynamics characteristics of the vehicle itself.

[0009] The motion sickness simulation robot is used as the test subject, compared with human body test, the robot can accurately reproduce the same test posture and perception state, and there is no score fluctuation caused by fatigue, emotion, and individual tolerance difference of the human body; At the same time, the robot can stably collect multi-dimensional physical stimulation data (such as acceleration, angular velocity, vibration frequency, etc.), and the data consistency is far superior to the subjective score of the human body, so that the evaluation result has reproducibility.

[0010] Repeated testing and data mean processing: through repeated testing at the same position several times and taking the arithmetic mean, the influence of random error (such as instantaneous fluctuation of sensor, slight deviation of working condition) on the physical stimulation vector is further reduced, and the data input into the prediction model accurately reflects the stable physical stimulation level of the vehicle at the position.

[0011] The method constructs a multi-dimensional physical stimulation vector (which can cover longitudinal / lateral / vertical acceleration, angular acceleration, vibration frequency, duration, etc. Key parameters), which comprehensively covers the core physical causes of motion sickness (visual-vestibular mismatch, body vibration stimulation, etc.), and the evaluation dimension is more in line with the physiological mechanism of motion sickness.

[0012] When the predicted motion sickness score exceeds the threshold, the key physical stimulus components that cause the threshold exceed (such as excessive lateral acceleration fluctuations, vertical vibration frequencies concentrated in the 0.5-2Hz sensitive interval) can be directly located, breaking through the limitations of traditional evaluation that only knows unqualified but does not know why it is unqualified, and providing a clear direction for vehicle optimization (such as adjusting suspension damping and optimizing power response curve).

[0013] Quantitative indicators support engineering optimization: the quantitative correspondence between multi-dimensional physical stimulus vectors and motion sickness scores enables the conversion of vehicle motion sickness comfort from subjective perception to quantifiable and optimizable engineering indicators, facilitating targeted improvements by vehicle manufacturers during the research and development stage (such as adjusting chassis parameters through simulation and optimizing acceleration and deceleration strategies of driving assistance systems), thereby improving optimization efficiency.

[0014] As a preferred embodiment, the motion simulation robot is equipped with a three-axis accelerometer, a three-axis gyroscope, a three-axis jerk meter, a vibration pickup, a sound level meter, a CO2 / O2 concentration sensor, a wind speed meter, and a line-of-sight jitter camera to synchronously collect X / Y / Z linear acceleration, angular velocity, jerk, vibration frequency, noise, CO2 concentration, O2 concentration, field angular velocity, and line-of-sight jitter amplitude at a frequency of 1 kHz.

[0015] As a preferred embodiment, the comfort threshold score is calibrated through prior human factor tests, including the following: Select ≥80 volunteers covering 18-59 years old, half male and half female, and half high / low MSSQ scores, complete 46 minutes of riding in the same vehicle under the same test cycle, and record real-time subjective motion sickness scores using an 11-level MISC scale; perform random forest regression on the peak MISC of the volunteers and the synchronously collected multi-dimensional physical stimulus vectors of the robot, and take the model output value corresponding to MISC=4 as the comfort threshold score.

[0016] As a preferred embodiment, the multi-dimensional physical stimulus vector P is sequentially executed before inputting the model M: Outlier rejection: remove outliers using the 3σ criterion for any dimension; Time alignment: resample the data of each sensor to 1 kHz based on the gyroscope timestamp; Standardization: perform Z-score normalization for each dimension to make the mean value 0 and the variance 1.

[0017] As a preferred embodiment, the motion sickness prediction model is established, including the following: Based on robot parameters: calculate the Pearson and Spearman correlation coefficients between the robot acquisition parameters and the subjective score, screen the key parameters, and use a multiple linear regression model, with the subjective car sickness score as the dependent variable Y and the key physiological indicators as the independent variables , , … , the model formula is:

[0018] wherein is the intercept term, , , … is the regression coefficient, is the error term; A mapping model is established, and the cross-validation MSE, MAE, and R² are evaluated.

[0019] As an implementable preferred solution, the driver training test includes the following contents: At the same entrance and exit point of the performance road in the test field, the driver completes 5 laps of continuous circulation, and the proportion of the period in which the speed-acceleration deviation exceeds ±5% is calculated as the overage rate; only when the overage rate of the last 3 consecutive training is ≤5%, is the driver determined to have formal test qualifications.

[0020] As an implementable preferred solution, the SHAP value analysis algorithm is used to calculate the contribution rate of each physical stimulation component to the motion sickness score; If the top 3 items with the highest contribution rate include longitudinal jerk or Z-axis 4-12 Hz vibration, it is recommended to adjust the motor torque filter, energy recovery intensity, or suspension damping; if CO2 concentration is included, it is recommended to improve the air exchange amount in the vehicle; the optimized vehicle is re-executed for the test step until the motion sickness score is ≤ the comfortable threshold score.

[0021] In a second aspect, the present application also provides a motion sickness comfort evaluation system for an electric passenger vehicle, characterized in that the above-mentioned motion sickness comfort evaluation method for an electric passenger vehicle is used. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a logic diagram of a motion sickness comfort evaluation method for an electric passenger vehicle.

[0023] Figure 2 is a test working condition diagram. DETAILED DESCRIPTION

[0024] In order to make the technical solutions of the present application and the advantages thereof clearer, the technical solutions of the present application will be further described in detail below in conjunction with the drawings. It can be understood that the specific embodiments described herein are only partial embodiments of the present application, and are only used to explain the present application, but not to limit the present application. It should be noted that the technical features described in the following embodiments or the combination of technical features should not be considered as isolated, and they can be combined with each other to achieve better technical effects. The same reference numerals appearing in the drawings of the following embodiments represent the same features or components, which can be applied to different embodiments.

[0025] In addition, unless otherwise defined, the technical terms or scientific terms used in the description of the present application should have the usual meanings understood by those of ordinary skill in the art to which the present application belongs.

[0026] The present application will be further described in detail below in conjunction with the drawings.

[0027] Embodiment one With reference to Figure 1 The present disclosure provides a method for evaluating the comfort of motion sickness of an electric passenger car, comprising the following steps.

[0028] Step S100, test condition design, with reference to Figure 2 , comprising: Step S101, based on user real scene design, mainly simulating urban road driving conditions, including acceleration, deceleration, lane changing, overtaking, curve, deceleration zone and other conditions. The speed points and throttle brake pedal opening degrees in the working condition are designed based on the surrounding city actual road exploration data and the relevant data of the China city commuting detection report, and the working condition design parameters of other detection institutions are referred to Step S102, design using test field, including: selecting test field performance road for working condition design, the starting point and the ending point are both at the entrance of the performance road, each circle is one driving cycle, each group of volunteers drives 5 cycles in human factor test, a total of 46 minutes.

[0029] Step S103, environmental condition setting, the test road is dry and smooth asphalt road, wind speed < 5 m / s, in order to reduce the interference of environmental factors on the test results.

[0030] Step S104, repeatability guarantee measures, including: Driver selection: select a fixed driver, require driving age > 10 years, with smooth driving style.

[0031] Working condition prompting software: based on the actual measurement data of IMC gyroscope, integrating working condition prompting, throttle, brake pedal prompting, driving compliance out-of-tolerance rate prompting and background data statistical function.

[0032] The driver repeatedly practices each vehicle until the overage rate is qualified for three consecutive times, which is considered as the qualification of the driver. For example, during the practice, the software displays the deviation of the driving speed, acceleration and other parameters of the driver from the preset working condition in real time, and prompts when the deviation exceeds a certain range. The driver adjusts the driving operation according to the prompt until the overage rate of three consecutive driving meets the requirements.

[0033] Step S200, sample vehicle confirmation, according to the compact, medium, large classification, electric vehicle (pure electric + plug-in hybrid), fuel vehicle type, car, SUV type, according to the market sales of each level of car to determine the specific car model.

[0034] The number of sample vehicles should ensure the representativeness and universality of the test results.

[0035] Step S300, human factor test, including: Step S301, volunteer screening, including: Gender ratio: 1:1 for men and women to ensure gender balance of the sample.

[0036] Age ratio: 1:1 for young people (18-44 years old) and middle-aged people (45-59 years old) to cover different age groups.

[0037] Health condition: Volunteers have no cardiovascular / neurological diseases to ensure that physical condition will not interfere with test results.

[0038] Motion sickness history: distinguish between sensitive and non-sensitive groups, and use the Motion Sickness Susceptibility Questionnaire (MSSQ) to investigate the motion sickness susceptibility of volunteers.

[0039] Referring to Table 1, the MSSQ hot spring part contains a series of questions related to motion sickness experience, such as the frequency and severity of motion sickness when riding different vehicles, and the motion sickness susceptibility of volunteers is determined according to the questionnaire score.

[0040] Table 1

[0041] At the same time, apply for medical ethics to ensure that the test meets the ethical standards.

[0042] Step S302, subjective feeling index collection, including: Use the MISC scale to investigate the subjective feeling of motion sickness of volunteers in human factor test.

[0043] The MISC scale contains multiple level options to describe the symptoms of motion sickness, such as the severity of dizziness, which is divided into none, mild, moderate, severe, etc. Volunteers choose the appropriate level according to their own feelings.

[0044] The MISC scale in the experiment is as follows: The current feeling

single choice question

[0045] Step S303, objective physiological index collection, including: Collecting objective physiological indexes such as electroencephalogram, skin electricity, heart rate, etc. Professional physiological equipment is used to read the physiological indexes of the volunteers during the test, for example, the electroencephalogram equipment can monitor the brain wave changes of the volunteers in real time, the skin electricity equipment can measure the skin conductivity to reflect the stress state of the human body, and the heart rate equipment can accurately record the heart rate changes of the volunteers.

[0046] Step S304, test matrix design, two volunteers sit in each car, one sits in the co-driver seat and the other sits in the right rear seat. Each volunteer completes three state riding tests, and after completing three tests, the two test volunteers change positions and repeat the three state tests. The test order of 6 tests is generated by random number to reduce the influence of test order on the results.

[0047] Step S305, test process, including: Preparation: Vehicle preparation: install gyroscope for measuring vehicle acceleration, angular velocity and other motion parameters; install throttle pedal pull wire displacement and brake pedal pull wire displacement sensors to accurately record the driver's throttle and brake operation; install work condition prompt software tablet to provide work condition prompt information for the driver; install 6-DOF sensors at 4 positions (front, back, left and right) to comprehensively perceive the vibration and motion state of the vehicle; install cameras (preferably motion cameras) on the right front door and rear door to record the situation inside the vehicle.

[0048] Driver preparation: the driver repeatedly practices until the overage rate is qualified for 3 consecutive times, which is considered as the qualification of the driver.

[0049] Volunteer preparation: install and debug the electroencephalogram equipment to ensure that the electroencephalogram signals can be accurately collected; wear a physiological bracelet for the volunteer to monitor the heart rate and other physiological indicators in real time; place a portable speed bump at the north bend of the outermost lane of the performance road to simulate the speed reduction scene in the actual road.

[0050] Test process: Seating arrangement: one volunteer sits in the front passenger seat, one volunteer sits in the rear right seat, and one test auxiliary personnel sits in the rear left seat to ensure the safety of the test and the smooth progress of data collection.

[0051] Device start and setting: start the gyroscope and work condition prompt software after driving the vehicle to the starting point of the performance road entrance.

[0052] Set the work condition prompt software, including device connection, pull wire displacement initial position zero, 100% stroke setting, etc., and confirm whether the report page is displayed normally.

[0053] Start the camera to ensure that it is recording normally.

[0054] Click the start key on the software interface, select the number of cycles (5 times in this embodiment), set the driver's name (such as CAERI-20250817-1, where 1 is the test number for the day), and set the driver's name.

[0055] Start the test: drive the vehicle according to the prompts of the work condition prompt software, with each lap as a cycle. Stop the vehicle during driving to conduct a subjective perception questionnaire survey for the volunteers. Each test needs to be conducted continuously for 5 cycles, a total of 46 minutes.

[0056] Test completion: after the test is completed, click Save PDF on the report interface of the work condition prompt software and name it with the date and test number for the day.

[0057] Step S306, human factor test data processing, at least 8 volunteers for each vehicle, forming three seating states, with 15 groups horizontally and 8 groups vertically for each seating state.

[0058] Data cleaning is performed to eliminate invalid data, such as abnormal data caused by equipment failure or improper operation of volunteers; time synchronization of each variable is ensured, and data collected by different devices are aligned according to the time axis; and data are standardized to make data of different dimensions comparable.

[0059] Step S400, motion sickness simulation robot test, comprising: The motion sickness simulation robot is placed in the co-pilot and rear right position of each test vehicle, and the same driver performs the motion sickness test under the same working conditions.

[0060] During the test, the motion sickness simulation robot collects the following data: XYZ three-direction speed, acceleration, jerk, angular velocity, angular acceleration, vibration frequency, noise, carbon dioxide, oxygen, field of view, line of sight jitter, and wind speed. Each position is tested for 3 times, and the arithmetic mean of the final data is taken to reduce test errors.

[0061] Data cleaning is performed, including: outlier elimination: 3σ criterion is used to remove mutation points in any dimension; time alignment: the gyroscope timestamp is used as the reference to resample the sensor data to 1 kHz; standardization: Z-score normalization is performed on each dimension to make the mean value 0 and the variance 1, so as to eliminate the dimensional difference.

[0062] Step S500, model establishment, comprising: Motion sickness prediction model based on objective physiological information: Correlation analysis: calculate the correlation coefficient of each physiological parameter (such as electroencephalogram, skin electricity, heart rate, etc.) and subjective car sickness score. Pearson correlation coefficient (for linear correlation analysis) and Spearman rank correlation coefficient (for nonlinear correlation analysis) can be used.

[0063] For example, calculate the Pearson correlation coefficient between the power of a certain frequency band in the electroencephalogram signal and the subjective car sickness score to determine the degree of linear correlation.

[0064] Statistical / machine learning modeling: according to the correlation analysis results, filter the key physiological indicators and establish a quantitative mapping relationship. Multivariate linear regression (MLR), ridge regression / Lasso regression, random forest regression / gradient boosting tree, and support vector regression (SVR) methods can be used.

[0065] In this embodiment, a multivariate linear regression model is used, where the subjective car sickness score is the dependent variable , and the key physiological indicators are the independent variables , , … The model formula is:

[0066] where, is the intercept term, , , … is the regression coefficient, is the error term.

[0067] The regression coefficients are estimated by least squares method to minimize the error between the model predicted values and the actual values.

[0068] Model validation: cross-validation is used to evaluate the prediction accuracy, commonly used evaluation indicators include mean squared error (MSE), mean absolute error (MAE), coefficient of determination (R²) and so on.

[0069] For example, divide the dataset into k subsets, use k - 1 subsets for training each time, and the remaining 1 subset for validation, repeat k times, calculate the MSE, MAE and R² value of each validation, and take the average value as the final evaluation index of the model.

[0070] Based on the motion simulation robot parameters, the motion prediction model is established, Correlation analysis: calculate the correlation coefficient between each robot parameter (such as speed, acceleration, vibration frequency, etc.) and subjective car sickness score, also use Pearson correlation coefficient and Spearman rank correlation coefficient.

[0071] For example, analyze the Spearman rank correlation coefficient between vehicle acceleration and subjective car sickness score to determine the degree of non-linear correlation.

[0072] Statistical / machine learning modeling: select key robot parameters and establish quantitative mapping relationship, which can use the same methods as the motion prediction model based on objective physiological information, such as multiple linear regression, ridge / Lasso regression, random forest regression / gradient boosting tree, support vector regression, etc.

[0073] For example, use random forest regression model to construct multiple decision trees for comprehensive analysis of key robot parameters to predict subjective car sickness score.

[0074] Model validation: also use cross-validation to evaluate the prediction accuracy, calculate MSE, MAE, R² and other indicators to ensure the reliability and accuracy of the model.

[0075] Step S600, based on the motion simulation robot parameters, the motion test procedure includes: Equipment installation and debugging: Install gyroscope, throttle pedal pull line displacement, brake pedal pull line displacement, working condition prompt software tablet and other equipment, and debug each equipment to normal parameters.

[0076] For example, check the sampling frequency and measurement accuracy of the gyroscope to ensure that it can accurately measure the motion parameters of the vehicle; adjust the throttle and brake pedal pull wire displacement sensor to accurately record the driver's operation.

[0077] Working condition driving practice: the driver follows the working condition prompt software to perform the same test cycle until the consecutive three times of driving exceed the tolerance rate is qualified, ensuring that the driver can accurately drive according to the preset working condition.

[0078] Formal test: Install the robot in the co-pilot position and have the same driver perform the same 3 working condition drives, each separated by more than half an hour to avoid errors caused by continuous work of the robot.

[0079] Copy robot test parameters.

[0080] Install the robot in the co-pilot position and have the same driver perform the same 3 working condition drives, each separated by more than half an hour to avoid errors caused by continuous work of the robot.

[0081] Copy robot test parameters.

[0082] Data processing: based on the prediction model, data cleaning is performed on the strongly correlated parameters to eliminate invalid data and ensure time synchronization of each variable; data alignment; standardization processing.

[0083] Take the arithmetic mean of the three tests as the multi-dimensional physical stimulation vector of each seating position.

[0084] Test results: input the multi-dimensional physical stimulation vector of each seating position into the motion sickness prediction model based on the motion sickness simulation robot parameters established in the early stage, and output the motion sickness test results of each seating position.

[0085] Motion sickness optimization: combined with the threshold of the motion sickness prediction model, the optimization direction of this parameter is proposed. For example, if the model shows that the acceleration change of the vehicle is closely related to the occurrence of motion sickness, and when the acceleration change rate exceeds a certain threshold, the incidence of passenger motion sickness significantly increases, then it is recommended to optimize the control strategy of the vehicle's power system to reduce the acceleration mutation during acceleration and deceleration to reduce the incidence of passenger motion sickness.

[0086] In one embodiment, the SHAP value analysis algorithm is used to calculate the contribution rate of each physical stimulation component to S; If the top 3 items with the highest contribution rate include longitudinal jerk or Z-direction 4-12 Hz vibration, it is recommended to adjust the motor torque filter, energy recovery intensity or suspension damping first; if CO2 concentration is included, it is recommended to improve the air exchange rate in the vehicle; The optimized vehicle is re-executed for the test step until S≤S0, and the motion sickness comfort closed-loop verification is completed.

[0087] Embodiment Two The present application adopts a two-stage structure combining a teacher model containing objective physiological signals with a student model without objective physiological signals, and introduces a knowledge distillation method, so that the student model can still obtain motion sickness evaluation capability close to the teacher model without relying on physiological signals.

[0088] Step S1, teacher model construction (contains physiological signal input): The teacher model is used to learn the mapping relationship between "motion parameters + environmental parameters + physiological signals" and "subjective motion sickness level". To support the distillation process, the teacher model adopts a neural network structure that can output intermediate feature layers and temperature prediction results.

[0089] Network structure: The teacher model adopts a branch type time series neural network, and uses the subjective score sequence as the supervision target during training. It includes: Motion parameter branch: one-dimensional convolution layer or time series attention layer is used to extract the dynamic features of vehicle and dummy motion data; Physiological signal branch: convolution layer or bidirectional gated recurrent layer is used to extract the change pattern of EEG, GSR, HR, pupil diameter and other signals; Environmental parameter branch: uses a fully connected layer to model the stable trend of environmental variables; Fusion layer: concatenates the three types of features in the channel dimension, and uses several residual blocks for fusion; Output layer: outputs continuous motion sickness level prediction value, and outputs probability distribution of each level, which can be adjusted through temperature parameter.

[0090] Teacher model training: Masking, normalization and time window slicing are performed on each input modality. The mean absolute error or cross-entropy is used as the loss function, and the Adam optimizer is used for training. After training, the following information is retained for distillation: Soft prediction of the teacher model (temperature probability or continuous output); Key intermediate feature layer output of the teacher model; Vector representation of the feature fusion layer.

[0091] Step S2, student model construction (only relies on motion and environmental input): The student model is used for actual deployment, and its input does not contain any physiological signals, but only consists of vehicle and dummy motion parameters and environmental parameters.

[0092] Network structure: The student model adopts a time sequence neural network similar to the teacher model structure but lighter, including a motion feature extraction module, an environment feature extraction module and a fusion prediction module. The model capacity is controlled at about 20% of the teacher model, improving the inference speed and deployment feasibility.

[0093] Training target The student model receives the following in the training phase: Subjective evaluation hard labels (MISC, SSQ, etc.); Soft targets output by the teacher model; Intermediate layer features in the teacher model.

[0094] The student model needs to learn the above information through the distillation framework to approximate the decision-making behavior of the teacher model.

[0095] Step S3, multi-level knowledge distillation mechanism: To effectively transfer the physiological signal information contained in the teacher model to the student model, the present application constructs a three-layer distillation mechanism, including: Soft target distillation: Soft target distillation can make the fine-grained relationship between the motion sickness levels contained in the teacher model output be learned by the student model through temperature processing.

[0096] Teacher model output:

[0097] Student model output:

[0098] Distillation loss:

[0099] Feature distillation: To preserve the dynamic feature patterns formed by physiological signals in the teacher model, the present application takes the key feature layer of the teacher model as a supervision signal.

[0100] Let the intermediate feature vectors of the teacher model and the student model be , , then the feature distillation loss is:

[0101] This part makes the student model approximate the teacher model in the internal feature space, so that it can still learn the equivalent change trend without physiological input.

[0102] Distillation representation: Teacher model fusion representation is the feature representation produced by the fusion layer (or bottleneck layer) before the output head of the teacher model, which integrates the three types of information: motion parameters, physiological signals, and environmental parameters. is the feature representation produced by the fusion layer before the output head of the student model, which is only composed of motion parameters and environmental parameters.

[0103] The distillation loss is represented as:

[0104] This mechanism ensures that the final hidden representation space of the student model without physiological signals is consistent with the teacher model, and its output geometry and confidence distribution are closer to the teacher model, making it easier for subsequent temperature calibration or equivalent regression to converge. When combined with SHAP and other methods for analysis, the input features have clearer paths to the representation and output.

[0105] Step S4, joint training loss function: The training goal of the student model is composed of the following three parts:

[0106] wherein is the supervision loss of the passenger's subjective evaluation label, , , , determined by the validation set.

[0107] The present application constructs a teacher model based on vehicle and dummy motion parameters, environmental parameters, and passenger physiological signals, and through multi-source data fusion and neural network model training, the motion sickness level is converted from traditional subjective description to quantifiable prediction output. The model is trained under a unified feature system and can reflect the continuous change trend of motion sickness degree, and the evaluation result has higher scientificity and reliability.

[0108] The present application collects data under standardized conditions and maintains the consistency of test inputs through automatic driving control. The student model after knowledge distillation can maintain stable discrimination ability only relying on motion and environmental parameters, making the evaluation of different vehicles and different time periods have high repeatability and comparability.

[0109] The student model constructed by the present application no longer needs to collect passenger physiological signals and does not need to rely on a large number of real human subjects. Motion sickness prediction can be completed only through dummy motion parameters and environmental parameters, thereby reducing the cost of test organization, improving the speed of evaluation, and improving the operability of engineering application.

[0110] The application can keep the sensitivity of the student model to the change of the key motion parameters in the reasoning stage by the distillation mechanism of the feature layer and the fusion representation. In combination with the explanatory analysis of the model, the main dynamic factors causing the high level of the motion sickness can be identified, thereby providing a clear technical basis for the vehicle suspension tuning, the power system calibration, the energy recovery strategy optimization and the improvement of the vehicle comfort.

[0111] The embodiment of the present disclosure further provides a motion sickness comfort evaluation system for an electric passenger vehicle, which uses the motion sickness comfort evaluation method for the electric passenger vehicle.

[0112] The above is only an embodiment of the present application, and the common knowledge of specific structures and characteristics in the scheme is not described in detail. The ordinary skilled person in the art knows all the ordinary technical knowledge in the field of the present application before the filing date or the priority date, can know all the prior art in the field, and has the ability to apply conventional experimental means before that date. The ordinary skilled person in the art can improve and implement the present scheme based on the disclosure given in the present application, and some typical known structures or known systems should not be an obstacle for the ordinary skilled person in the art to implement the present application. It should be noted that, for those skilled in the art, without departing from the structure of the present application, a number of modifications and improvements can be made, which should also be considered as the protection scope of the present application. The scope of protection of the present application should be subject to the content of its claims, and the specific implementation mode in the specification can be used to explain the content of the claims.

Claims

1. A method for evaluating motion sickness comfort in electric passenger vehicles, characterized in that, include: The same fixed driver continuously drives the tested electric passenger vehicle to the target number of laps under preset cyclic conditions; The motion sickness simulation robot was fixed to the passenger seat and the right rear seat of the vehicle under test, respectively. Repeat the test steps several times for the same driver, the same vehicle under test, and the same cyclic operating conditions, and take the arithmetic mean of the collected data as the multidimensional physical stimulus vector at that location. The multidimensional physical stimulus vector is input into a pre-calibrated and cross-validated motion sickness prediction model. The motion sickness prediction model takes the robot's physical stimulus vector as input and motion sickness score as output. The motion sickness prediction model directly outputs the predicted motion sickness scores for the front passenger seat and the right rear seat. If the predicted motion sickness score is higher than the preset comfort threshold score, the tested electric passenger vehicle is deemed to have unqualified motion sickness comfort in that position, and the key physical stimulus component that caused the threshold to be exceeded is located in reverse.

2. The method for evaluating motion sickness comfort in an electric passenger vehicle according to claim 1, characterized in that, The motion simulation robot is equipped with a three-axis accelerometer, a three-axis gyroscope, a three-axis accelerometer, a vibration pickup, a sound level meter, a CO2 / O2 concentration sensor, an anemometer, and a gaze jitter camera. It synchronously collects X / Y / Z linear acceleration, angular velocity, jerk, vibration frequency, noise, CO2 concentration, O2 concentration, field of view angular velocity, and gaze jitter amplitude at a location at 1 kHz.

3. The method for evaluating motion sickness comfort in an electric passenger vehicle according to claim 1, characterized in that, The comfort threshold score is calibrated through prior human factors testing and includes the following: ≥80 volunteers aged 18-59, with equal numbers of men and women and equal numbers of high and low MSSQ scores, were selected to complete a 46-minute ride in the same test vehicle under the same cyclic conditions. Real-time subjective motion sickness scores were recorded using an 11-level MISC scale. Random forest regression was performed on the volunteers' peak MISC scores and the robot's multidimensional physical stimulus vectors collected simultaneously. The model output value corresponding to MISC=4 was taken as the comfort threshold score.

4. A method for evaluating motion sickness comfort in an electric passenger vehicle according to claim 1, characterized in that, The multidimensional physical stimulus vector P is processed sequentially before being input into the model M: Outlier removal: The 3σ criterion is used to remove mutation points in any dimension; Time alignment: Based on the gyroscope timestamp, the data from each sensor is resampled to 1 kHz; Standardization: Perform Z-score normalization on each dimension so that the mean is 0 and the variance is 1.

5. The method for evaluating motion sickness comfort in an electric passenger vehicle according to claim 1, characterized in that, Establish a motion sickness prediction model, including the following: Based on robot parameters: Pearson and Spearman correlation coefficients were calculated between robot-collected parameters and subjective ratings. After screening key parameters, a multiple linear regression model was used, with subjective motion sickness rating as the dependent variable Y and key physiological indicators as independent variables. , , … The model formula is: in For the intercept term, , , … For regression coefficients, This is the error term; Establish a mapping model and evaluate it using cross-validation MSE, MAE, and R².

6. The method for evaluating motion sickness comfort in an electric passenger vehicle according to claim 1, characterized in that, Driver training tests include the following: At the same entrance and exit point on the performance track of the test site, the driver completes 5 consecutive laps. The percentage of time when the speed-acceleration deviation exceeds ±5% in each lap is the out-of-range rate. Only when the out-of-range rate of the most recent 3 consecutive training sessions is ≤5%, the driver is deemed to be qualified for formal testing.

7. A method for evaluating motion sickness comfort in an electric passenger vehicle according to claim 1, characterized in that, The contribution rate of each physical stimulus component to the motion sickness score was calculated using the SHAP value parsing algorithm. If the top three contributing factors include longitudinal acceleration or Z-axis 4-12 Hz vibration, it is recommended to adjust the motor torque filter, energy recovery intensity, or suspension damping; if CO2 concentration is included, it is recommended to increase the in-vehicle ventilation; repeat the test steps for the optimized vehicle until the motion sickness score is ≤ comfort threshold score.

8. A motion sickness comfort assessment system for electric passenger vehicles, characterized in that, The method for evaluating motion sickness comfort in electric passenger vehicles as described in any one of claims 1-7 was employed.

Images