A vehicle ride comfort evaluation method and system

Abstract

The present application belongs to the technical field of intelligent transportation, and specifically discloses a vehicle ride comfort evaluation method and system. In view of the problems that the existing driving evaluation system relies on subjective evaluation of passengers, which is not objective enough, the anti-interference ability of single sensor data is poor, and the individual physical differences of passengers are ignored, the method of the present application fuses physiological sensor data of passengers, vehicle motion data and multi-modal spatio-temporal coupling features, which is conducive to accurately judging the driving aggressiveness of drivers, further combines the confidence probability of aggressive driving behavior, and calculates the single trip score based on the dynamic scoring algorithm of mileage normalization. At the same time, a time decay factor is introduced to evaluate the driver in the long term to obtain the comprehensive score of the driver, and to realize the objective evaluation of the vehicle ride comfort. In the scoring calculation process, the crowd sensing mechanism is introduced to correct the passenger sensitivity, which is finally conducive to improving the objective quantitative evaluation of the passenger ride comfort of the vehicle.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation technology, and specifically relates to a method and system for evaluating vehicle passenger comfort. Background Technology

[0002] Existing methods for evaluating driving comfort often rely on subjective passenger ratings or only consider driver behavior characteristics, potentially leading to contradictions where drivers perceive their driving as mild while passengers perceive it as aggressive. Furthermore, different passengers exhibit vastly different physiological response thresholds to vehicle motion stimuli (such as acceleration, deceleration, and turning). Sensitive passengers (e.g., those prone to motion sickness) may experience significant heart rate variability even with slight bumps, while less sensitive passengers may respond calmly to aggressive driving. Existing "one-size-fits-all" static threshold models fail to establish personalized baselines and correction coefficients based on historical passenger data, resulting in significant biases in driving comfort evaluations and reducing the fairness and generalization ability of the evaluation system.

[0003] It is evident that existing methods for evaluating driving comfort do not adequately consider passengers' physiological comfort and the differences in individual physical condition and physiological sensitivity. Furthermore, current technologies rely excessively on single-dimensional vehicle physical data or passenger physiological data for data collection and analysis. Physical vehicle indicators alone cannot detect "hidden aggressive" behaviors that cause passenger discomfort but do not trigger thresholds; physiological indicators alone lack constraints related to vehicle motion. This fragmented data analysis model fails to establish a causal loop of "physical stimulus-physiological response," often leading to significant interpretative bias and easily misattributing non-driving factors (such as passenger emotional fluctuations) to the driver's responsibility, resulting in inaccurate judgments.

[0004] Furthermore, existing technologies have significant limitations in signal processing capabilities under complex and dynamic road conditions, making it difficult to meet the data confidence requirements of commercial applications. Although wearable devices have become increasingly common, current solutions have failed to effectively address the motion artifact problem within moving vehicles, where wrist vibrations caused by road bumps are often misinterpreted as sudden heart rate changes. In addition, existing technologies lack intensity matching verification mechanisms for non-driving scenarios, resulting in extremely poor robustness to environmental noise and emotional interference. This high false alarm rate fails to provide a reliable technical foundation for driver credit profiling, leading to weak anti-interference capabilities and a high false alarm rate.

[0005] In summary, there is an urgent need in this field to propose a driving comfort evaluation technology that can deeply integrate vehicle physical characteristics and passenger physiological characteristics, and has the ability to resist noise and clear biases. This technology would bridge the gap between "single-time experience perception" and "long-term credit settlement," thereby meeting the comprehensive requirements of the next generation of intelligent mobility systems for objectivity, accuracy, and humanization.

[0006] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art. Summary of the Invention

[0007] The purpose of this invention is to propose a method for evaluating vehicle ride comfort. By integrating passenger physiological sensor data, vehicle motion data, and multimodal spatiotemporal coupling features, it facilitates accurate judgment of the driver's aggressiveness. Furthermore, by combining the confidence probability of aggressive driving behavior, a dynamic scoring algorithm based on mileage normalization is used to calculate a single-trip score. Simultaneously, a time decay factor is introduced to conduct a long-term evaluation of the driver, obtaining a comprehensive driver score and achieving an objective evaluation of vehicle ride comfort. During the calculation of both the single-trip and comprehensive driver scores, a collective intelligence perception mechanism is introduced to correct for passenger sensitivity, ultimately improving the objective quantitative evaluation of passenger ride comfort.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A method for evaluating vehicle ride comfort includes the following steps: Step 1. Simultaneously acquire passenger physiological data and vehicle motion data, then perform motion artifact detection and removal, and then perform preliminary extraction on the removed data to obtain passenger physiological data and vehicle motion data; Step 2. Identify the baseline values ​​of passenger physiological data when the vehicle is in a stable driving state, and perform Z-Score standardization on the real-time collected passenger physiological data based on the baseline values ​​to obtain normalized physiological features; Step 3. Extract feature variables from the standardized passenger physiological data and vehicle motion data; simultaneously extract two multimodal spatiotemporal coupling features, namely response delay time and physiological-physical energy coupling ratio; Step 4. Build an aggressive driving behavior recognition model. The input of the model includes the vehicle motion features, passenger physiological features and two multimodal spatiotemporal coupling features extracted in Step 3. The model output layer uses a Sigmoid function to map and output the confidence probability of aggressive driving behavior in the current time window; further, a dynamic scoring algorithm based on mileage normalization is executed based on the confidence probability to calculate the score for a single trip; Step 5. Use the cloud server to record the driver's historical trip data, introduce a time decay factor to conduct long-term evaluation of the driver, and obtain the driver's comprehensive score to evaluate the vehicle ride comfort. A collective intelligence perception mechanism is introduced into the process of single trip and driver comprehensive scoring to correct passenger sensitivity.

[0009] Furthermore, based on the aforementioned vehicle ride comfort evaluation method, this invention also proposes a corresponding vehicle ride comfort evaluation system, which adopts the following technical solution: A vehicle ride comfort evaluation system includes the following modules: The preprocessing module is used to simultaneously acquire passenger physiological data and vehicle motion data, then perform motion artifact detection and removal, and then perform preliminary extraction on the removed data to obtain passenger physiological data and vehicle motion data. The normalization module identifies the baseline value of passenger physiological data when the vehicle is driving smoothly, and performs Z-Score normalization on the real-time passenger physiological data based on the baseline value to obtain normalized physiological features. The feature extraction module is used to extract feature variables from the standardized passenger physiological data and vehicle motion data; it also extracts two multimodal spatiotemporal coupling features, namely response delay time and physiological-physical energy coupling ratio. The model building and single-trip scoring module is used to build an aggressive driving behavior recognition model. The model input includes extracted vehicle motion features, passenger physiological features, and two multimodal spatiotemporal coupled features. The model output layer uses a Sigmoid function to map and output the confidence probability of aggressive driving behavior in the current time window; further, a dynamic scoring algorithm based on mileage normalization is executed based on the confidence probability to calculate the score for a single trip; And a comprehensive scoring module, which uses cloud server to record the driver's historical trip data, introduces time decay factor to evaluate the driver over a long period of time, and obtains the driver's comprehensive score to evaluate the vehicle's ride comfort. A collective intelligence perception mechanism is introduced into the process of single trip and driver comprehensive scoring to correct passenger sensitivity.

[0010] The present invention has the following advantages: As described above, this invention discloses a method for evaluating vehicle ride comfort. This method introduces an IMU-based motion artifact suppression mechanism during the data processing stage, eliminating invalid physiological signals from non-stationary states at the source. Simultaneously, it combines dynamic baseline calibration technology to update passenger physiological baseline values ​​in real time during smooth driving. Z-score standardization eliminates measurement errors caused by individual passenger baseline heart rate differences or fatigue drift during long-distance travel, significantly improving the system's robustness in complex dynamic environments and exhibiting extremely high anti-interference capability and robustness. This invention innovatively constructs a spatiotemporal coupling feature of "physical stimulus-physiological response" and utilizes the XGBoost machine learning model to mine the nonlinear mapping relationship between multimodal features. This enables the invention to accurately determine aggressive driving events when "vehicle physical aggression" occurs and passenger physiological responses are within a reasonable delay and intensity range, effectively solving the technical problem that correlation does not equal causation, and improving event recognition accuracy. Furthermore, this invention proposes an objective and fair collective intelligence-based de-biased evaluation system. This system introduces a collective intelligence perception correction mechanism in the cloud, automatically calculating passenger sensitivity coefficients and weighting the scores by comparing individual historical responses with the group average. This collective intelligence perception mechanism smooths out score fluctuations caused by extreme passenger evaluations, ensuring the objectivity and fairness of driver evaluation results, reducing misjudgments due to overly sensitive passengers, and facilitating the objective quantitative evaluation of passenger comfort. Additionally, this invention proposes a time-sensitive dynamic credit incentive mechanism. This mechanism introduces a time decay factor that conforms to the laws of human memory, constructing a dynamically evolving driver credit profile. This mechanism assigns higher weight to recent trips, enabling the invention to both keenly capture drivers' recent risk tendencies and quickly reflect safety improvements after drivers improve their driving behavior, thereby promoting a virtuous cycle in the driving ecosystem. Attached Figure Description

[0011] Figure 1 This is an overall flowchart of the vehicle ride comfort evaluation method in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the acquisition and preprocessing process of passenger physiological data and vehicle motion data in an embodiment of the present invention; Figure 3 This is a schematic diagram of the multi-source data feature extraction process in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the process of splitting a driver's single trip equally in an embodiment of the present invention; Figure 5 This is a flowchart illustrating the driver's comprehensive score processing in an embodiment of the present invention. Detailed Implementation

[0012] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: Terminology Explanation: (1) Dynamic baseline calibration: Dynamic baseline calibration is an adaptive real-time signal processing mechanism designed to eliminate the interference of individual physiological differences and baseline drift over time on the assessment results. In multimodal physiological monitoring, the baseline indicators (such as heart rate variability) of different subjects are highly individual-specific and easily affected by factors such as long-distance fatigue, resulting in slow changes. Static thresholds are often difficult to guarantee accuracy. This mechanism uses real-time detection to dynamically calculate and extract the statistical characteristics of physiological signals during the period when the external physical environment is confirmed to be in a "non-disturbance state" such as when the vehicle is stationary or cruising at a constant speed, and updates it to the latest reference benchmark. Subsequently, using normalization algorithms such as Z-Score, the subsequent real-time absolute physiological values ​​are converted into "relative deviation" (i.e., surge) relative to the current benchmark.

[0013] This technology effectively isolates inherent physical biases and slowly varying background noise, ensuring that the features extracted by the system can objectively and accurately quantify the purely physiological stress response induced by transient external stimuli (such as aggressive driving events).

[0014] (2) Time decay factor: This is a mathematical mechanism used in time series analysis and dynamic credit assessment to adjust time weights. Based on the forgetting curve principle of human memory and behavioral habit evolution, it introduces a weight parameter that decreases with increasing time span to weight historical behavioral characteristic data. In long-term driving behavior evaluation systems, the traditional static arithmetic mean method cannot reflect the dynamic evolution characteristics of driving skills. The time decay factor assigns higher evaluation weights to drivers' recent trips and gradually diminishes the influence of long-term historical behavior, making the final comprehensive evaluation score highly timely.

[0015] This mechanism can not only keenly detect and warn of a driver’s recent risk propensity deterioration, but also quickly reflect the results of his improvement in safe driving habits, thus constructing a dynamic credit model that combines objective timeliness with positive incentives.

[0016] (3) Crowd Intelligence Perception: Crowd intelligence perception is a distributed sensing generic based on a large user group. It aims to achieve macro-level cognition and situational analysis of complex environments or behavioral characteristics by aggregating multi-source data collected from massive mobile edge nodes (such as smart wearable devices and smartphones) through collective intelligence collaboration. In the multimodal driving behavior evaluation system of this invention, the crowd intelligence perception mechanism is creatively used to construct a global physiological benchmark and individual sensitivity profile.

[0017] In this invention, the average physiological stress response of all passengers on the platform under typical physical stress scenarios is statistically analyzed through cloud big data analysis to obtain a statistically significant "group consensus baseline". Then, the historical physiological characteristics of a passenger are quantitatively compared with the obtained group baseline to dynamically calculate its sensitivity correction coefficient.

[0018] This technology utilizes the objective attributes of macro-level group data to effectively eliminate evaluation noise introduced by individual passengers' "easily startled" or different subjective cognitive thresholds. This achieves a logical closed loop of using collective intelligence to correct individual subjective biases, thereby ensuring the objectivity, scientific rigor, and fairness of the comprehensive credit settlement for drivers.

[0019] To address the shortcomings of existing driver evaluation systems, such as reliance on subjective passenger assessments lacking objectivity, poor anti-interference capabilities of single-sensor data, and neglect of individual passenger physical differences, this invention proposes a vehicle ride comfort evaluation method. This method effectively eliminates misjudgments caused by road bumps, passenger limb movements, or non-driving-related emotional fluctuations by integrating noise reduction technology from wearable device inertial sensors with spatiotemporal causal verification of vehicle motion data. It transforms passengers' implicit "physical discomfort" and "psychological stress" into quantifiable objective physiological indicators, filling the gap in traditional vehicle physical sensors' inability to perceive ride comfort. Furthermore, this invention utilizes cloud-based big data to construct passenger physiological sensitivity profiles and automatically corrects scoring biases caused by individual passenger physical differences (such as susceptibility to motion sickness or panic) through statistical debiasing algorithms. This eliminates the occasional impact of a single trip, providing drivers with a fairer, more stable, and long-term reliable credit profile.

[0020] like Figure 1 As shown, the vehicle ride comfort evaluation method in this embodiment includes the following steps: Step 1. Simultaneously acquire passenger physiological data and vehicle motion data, then perform motion artifact detection and removal, and finally perform preliminary extraction on the removed data to obtain passenger physiological data and vehicle motion data, such as... Figure 2 As shown.

[0021] First, it is necessary to obtain synchronized and reliable initial feature data through wearable devices, vehicle sensors, and GPS.

[0022] The wearable device used in this invention for collecting passenger physiological data is, for example, a smartwatch, but it can also be replaced with other wearable devices that can obtain physiological data, such as smart bracelets or smart rings, depending on the conditions.

[0023] The wearable device collects the passenger's heartbeat and pulse signals through a built-in PPG (photoplethysmography) sensor; and monitors the spatial motion state (acceleration, angular velocity) of the wrist or limb in real time through a built-in IMU (inertial measurement unit).

[0024] The device used to acquire vehicle motion data is an in-vehicle inertial navigation system, while the device used to acquire vehicle location information is GPS. Real-time vehicle location information is obtained via the mobile phone's built-in GPS, and vehicle motion data is acquired via the in-vehicle inertial navigation system.

[0025] Data acquired by smartwatches, in-vehicle inertial navigation systems, and GPS is transmitted in real time to smart mobile terminals (such as mobile phones).

[0026] The smart mobile terminal collects multi-source heterogeneous data from these sources, processes the data, and feeds it into a pre-trained model deployed on the smart mobile terminal. The model then provides a real-time evaluation of a single trip.

[0027] After the trip concludes, the data will be uploaded to the platform's data center (i.e., the cloud). The cloud is responsible for storing massive amounts of historical data, mining macro-level patterns within the group, and facilitating long-term settlement of dynamic credit.

[0028] However, two core issues still need to be addressed in this step: First, there is a time synchronization issue between smartwatches, in-vehicle inertial navigation systems, GPS, and mobile phones; second, passenger body movements may cause motion artifacts on the photoplethysmography (PPG) signal. Therefore, the following steps are designed to address these issues.

[0029] Step 1.1. Multi-device clock synchronization and frequency alignment.

[0030] Passenger HRV physiological data is extracted using wearable devices, and vehicle motion data is simultaneously acquired using in-vehicle inertial navigation and GPS while acquiring passenger physiological data.

[0031] Because there are certain discrepancies in the timestamp positioning between the smartwatches worn by passengers, in-vehicle inertial navigation, GPS and mobile terminals, and the initial data sampling rates may not be the same, spatiotemporal alignment is necessary.

[0032] The soft clock synchronization mechanism employs an application-layer synchronization protocol based on the NTP (Network Time Protocol) principle. The mobile terminal acts as the "master clock," while smartwatches and in-vehicle inertial navigation systems serve as "slave clocks." Synchronization process: 1) The mobile terminal sends a synchronization command and records the sending time. .

[0033] 2) Smartwatches and in-vehicle inertial navigation systems receive commands and record the reception time. .

[0034] 3) After processing, the smartwatch and in-vehicle inertial navigation system send a receipt and record the sending time. .

[0035] 4) The mobile terminal receives the receipt and records the reception time. .

[0036] Clock offset calculation formula: Calculate the time offset of the wearable device relative to the mobile terminal. The announcement is as follows: .

[0037] Network round-trip latency The formula is expressed as follows: .

[0038] The offset is applied only when δ < 50ms. Timestamps of all data collected by wearable devices Make corrections: .

[0039] in, This is the corrected timestamp.

[0040] At the same time, since different sensors have different sampling rates, frequency alignment is required.

[0041] This invention uses a high-frequency 50Hz standard for alignment and employs a linear interpolation algorithm to fill in missing time points for low-frequency data, so that it achieves a point-to-point correspondence with physiological data on the time axis.

[0042] Step 1.2. Motion artifact detection and signal cleaning.

[0043] Based on wrist inertial sensor data continuously collected by wearable devices, the intensity of movement within a local time window is calculated to identify and eliminate motion artifacts in physiological signals caused by the passenger's own limb movements.

[0044] This step is an important part of edge computing. Its purpose is to prevent subsequent models from misinterpreting a passenger's "waving," "looking up at their watch," or "sudden shaking" as a spike in heart rate. Therefore, the data measured by inertial navigation needs to be filtered.

[0045] I. Calculation of wrist composite acceleration.

[0046] Collect three-axis acceleration data from wearable devices , , ; Calculate the magnitude of the combined wrist acceleration: ; in This represents the instantaneous composite acceleration of the wrist at time t.

[0047] II. Dynamic signal separation.

[0048] A robust sliding window variance method is used, and the window size is set. The formula for calculating the standard deviation of wrist acceleration is as follows: ; in It is the mean within the window. This represents the standard deviation of acceleration measured within the sliding time window.

[0049] III. Motion artifact detection.

[0050] Set motion interference threshold Motion artifact mask discrimination formula: ; in This represents the signal confidence mask.

[0051] when When the confidence score is 0, the current physiological signal is determined to be contaminated by motion noise, and the data for that period is discarded directly, or its corresponding confidence weight is set to 0 and it is not included in the calculation of subsequent steps.

[0052] when When = 1, this segment of physiological data is retained for heart rate extraction.

[0053] Step 1.3. Data extraction.

[0054] The discriminated data is initially extracted to obtain physiological data, including HRV and EDA, as well as vehicle motion data, including longitudinal velocity, acceleration, jerk and vehicle position information, which are required for subsequent model input.

[0055] After data cleaning and alignment, the data extracted from smartwatches and in-vehicle inertial navigation systems has a preliminary level of reliability. Based on this, preliminary data extraction is then performed. The extracted data includes: I. Vehicle motion data (collected via onboard inertial navigation).

[0056] Instantaneous velocity; It is longitudinal acceleration; For accelerometer ( (differential); in, .

[0057] II. Passenger physiological data (collected via smartwatch).

[0058] Through the processing in step 1 above, we can finally obtain time-synchronized passenger physiological data and vehicle motion data that have been filtered out of invalid data, laying a solid foundation for the subsequent input model to achieve the purpose of the invention.

[0059] Step 2. Identify the baseline values ​​of passenger physiological data when the vehicle is in a stable driving state, and perform Z-Score standardization on the real-time collected passenger physiological data to obtain normalized physiological characteristics, such as... Figure 3 As shown.

[0060] Based on the preliminary data collected in step 1, due to the huge differences in the basic physiological indicators and change patterns of different individual passengers, and the fact that the physiological indicators of the same passenger may drift due to fatigue during a long journey.

[0061] Therefore, it is necessary to obtain the physiological indicators of passengers in the calm driving range as a baseline for subsequent changes in passenger indicators.

[0062] Specifically, the vehicle's driving status is identified in real time. When the vehicle is detected to be in a stable driving or stationary state, the passenger's physiological baseline parameters (i.e., the values ​​of indicators such as the passenger's heart rate variability in a stable state) are updated using the physiological data of the current time period. The real-time collected physiological data is then subjected to Z-Score standardization to obtain normalized physiological data.

[0063] Step 2.1. Monitor vehicle status using the collected vehicle motion data and GPS data, and define a time period that meets the following conditions as a stable window: longitudinal acceleration accelerometer Duration .

[0064] Step 2.2. Within the stationary window, calculate the moving average of the passenger HRV (heart rate variability) index. and standard deviation data .

[0065] Step 2.3. Perform real-time standardization using Z-score, utilizing the moving average of the passenger HRV index calculated within the stationary window. and standard deviation data For subsequent moments Real-time HRV value Perform the conversion.

[0066] Z-score (standard score) is a statistical normalization method used to measure the distance (in standard deviations) of a data point from the mean. The Z-score formula is: .

[0067] in This refers to the standardized numerical values. In this invention, physiological indicators are converted from "absolute values" to "the degree of deviation relative to the current resting state," thereby obtaining normalized physiological characteristics and achieving objective evaluation across individuals.

[0068] Step 3. Extract feature variables from the standardized passenger physiological data and vehicle motion data. Simultaneously extract two multimodal spatiotemporal coupling features: response delay time and physiological-physical energy coupling ratio, such as... Figure 3 As shown.

[0069] After completing the first two steps of bias correction, dynamic baseline calibration, and standardization, the passenger physiological data and vehicle motion data were processed to extract feature variables. A total of 24 HRV features (i.e., heart rate variability) and 10 vehicle motion features were extracted.

[0070] The extracted features are shown in Tables 1 and 2 below.

[0071] Table 1 HRV characteristics

[0072] Table 2 Vehicle Motion Characteristics

[0073] The data, after undergoing motion artifact detection, dynamic baseline calibration, and real-time Z-Score standardization in the above steps, were used for feature extraction. A total of 24 HRV features and 10 vehicle motion features were extracted.

[0074] Based on the extraction of the commonly used feature variables mentioned above, this invention also extracts spatiotemporal coupling features to describe the interaction between physical stimuli and physiological responses. The specific extraction process is as follows: I. Response latency , used to describe the time difference between the occurrence of a physical stimulus and the peak time of a physiological response; .

[0075] in The value is the Z-Score standardized value of the HRV calculated in real time in step 2. Extract the calculated jerk from step 1; This indicates the time corresponding to when the maximum value is reached.

[0076] The physiological response of the human body to fright is usually delayed by 0.5s to 3s. If the response delay τ is within this range, it is judged as a real radical event. If τ < 0, that is, the reaction occurs before braking, or τ > 10s, it is judged as noise.

[0077] II. Physiological-physical energy coupling ratio This is used to measure whether the "stimulus" and "response" match, and the formula is as follows: .

[0078] Where ε is a small constant to prevent the denominator from being zero (e.g., 1e-6); and Both are scalar values. This represents the quantitative value of the total "psychological stress" experienced by passengers during this incident.

[0079] Based on the calculation obtained in the previous step The sequence is used to integral the square of the passenger's stress level within a response time window following the physical event (typically delayed by a few seconds), while also introducing... The function, which is the squared energy integral of the positive deviation representing the physiological stress state, is given by the following formula: .

[0080] A quantitative value representing the "total physical impact energy" released during a sudden braking or sharp turn.

[0081] based on Sequence, setting aggressive trigger threshold ,when > When a suspected aggressive driving incident is identified, a time window of the incident is captured, and the square integral of the acceleration within that time window is performed, as shown in the following formula: .

[0082] for and Time window for physical energy calculation It is based on the absolute value of the instantaneous acceleration first exceeding the radical trigger threshold. Open the time window at that time.

[0083] Until Falling back to the safe exit threshold And the time window is closed after maintaining the preset duration of 1 second; in The moment when the event begins, i.e. The time is identified as the start time of the event. Start energy integration; The end time of the event, i.e. when When, mark as the end time of the event. Stop integrating.

[0084] In this embodiment Set at 2.0 m / s 3 ~ 3.0 m / s 3 ; =0.5m / s 3 .

[0085] Physiological-physical energy coupling ratio The physical meaning of this feature is to reflect the degree to which passengers respond to stimuli.

[0086] like A moderate value indicates a normal aggressive driving reaction; if Extremely high values ​​indicate that the passenger has a strong reaction to stimulation or has been subjected to other stimuli; if An extremely low value indicates that the passenger is slow to react when stimulated.

[0087] Solving for the above two parameters is to break through the fragmentation of traditional technology: either only looking at the car (such as deducting points for sudden braking) or only looking at the person (such as alarming when the heart rate exceeds the standard), but this will lead to a lot of misjudgments (correlation is not the same as causation).

[0088] Set response delay time and physiological-physical energy coupling ratio These two parameters, from the perspectives of time and intensity, demonstrate that the passengers' current panic is indeed caused by the vehicle's unstable operation just now.

[0089] These two parameters are used in the inference process of the XGBoost multimodal fusion decision algorithm. During the growth of the decision tree, the XGBoost model automatically finds the split point based on the two coupled features and adaptively generates decision logic. It is precisely because of the participation of these two parameters that the XGBoost model can use a greedy algorithm to traverse and optimize to accurately evaluate the information gain of the feature split point, adaptively establishing these two spatiotemporally coupled features with strong causal properties as the core split nodes of the decision tree. This makes these two parameters the key to comprehensively judging whether the stimulus and response match, thus ensuring the subsequent output of an accurate probability P of aggressive driving confidence.

[0090] Step 4. Build an aggressive driving behavior recognition model based on XGBoost. The input feature matrix X contains the vehicle motion features, passenger physiological features, and two multimodal spatiotemporal coupling features extracted in Step 3.

[0091] The aggressive driving behavior recognition model uses multiple gradient boosting trees for joint inference, and finally outputs the confidence probability of aggressive driving behavior existing in the current time window through a sigmoid function mapping at the output layer. Figure 4 As shown.

[0092] Specifically, the model will automatically start from the root node, sequentially select features, calculate the contribution of each feature to the current classification, and select spatiotemporal coupling features and other feature points with the largest gains for data partitioning. Then, by fitting the residual of the current model's prediction error in each iteration, the model's predictive ability is gradually improved, bringing the model closer to the optimal solution.

[0093] The output layer uses a Sigmoid function to map and output the confidence probability of aggressive driving behavior in the current time window (i.e., the credibility of each identified aggressive driving event). Finally, a single-time score is calculated based on the confidence probability of aggressive driving.

[0094] This invention uses XGBoost as an aggressive driving behavior classification model, which maps the preprocessed passenger physiological indicators, vehicle kinematic features, and spatiotemporal coupling features to a high-dimensional feature space.

[0095] The XGBoost model calculates first- and second-order gradient statistics and uses a greedy algorithm to traverse and optimize to accurately evaluate the information gain of feature split points. This adaptively establishes these two spatiotemporally coupled features with strong causal properties as the core splitting nodes of the decision tree, automatically dividing complex nonlinear decision boundaries. It can keenly capture the complex dynamic relationship between passenger physiological changes and vehicle motion, thus accurately judging aggressive driving events.

[0096] Meanwhile, the XGBoost model boasts advantages such as high accuracy, strong interpretability, and high flexibility, enabling it to solve problems quickly and accurately when handling large-scale datasets. Tree growth can be further controlled by setting parameters such as maximum tree depth (max_depth) and minimum number of samples. Regularization mechanisms ensure that XGBoost maintains its generalization ability when dealing with complex high-dimensional data, avoiding overfitting while improving its ability to distinguish between minority classes.

[0097] The model output layer outputs a simple binary judgment of whether or not an aggressive driving event occurs. Therefore, the Sigmoid function is needed to map the original prediction value of XGBoost to the aggressive driving confidence probability P (i.e. the credibility of each identified aggressive driving event), accurately quantifying the mathematical certainty of the aggressive driving event and providing a basis for subsequent single-trip scoring.

[0098] Of course, the aggressive driving behavior classification model in this embodiment can also be built based on LightGBM, CatBoost, temporal deep learning LSTM, GRU, Transformer, Attention, graph neural network GNN, etc.

[0099] After obtaining the confidence levels of each aggressive driving event identified by the model, dynamic scoring based on risk density is performed.

[0100] First, because multiple aggressive driving behaviors may occur during the trip, the system abandons the traditional discrete point deduction method and instead uses a crowd-sensing mechanism to calculate a risk index for each valid aggressive event. .

[0101] Among them, the risk index Defined as passenger sensitivity coefficient recorded in the cloud Confidence probability of model output With the physical energy of the event The product of the logarithmic functions is calculated using the following formula: .

[0102] Including the introduction of a risk index The aim is to smoothly compress the energy of extreme physical impacts, preventing a single extreme event from causing a collapse in the final score. The probability P is used to achieve severe punishment for "high-certainty and high-intensity" dangerous behaviors, and mild punishment for "low-certainty or low-intensity" behaviors; a sensitivity coefficient is introduced. The core objective is to eliminate evaluation biases caused by individual subjective irritability or insensitivity in advance within the bottom-level deduction unit.

[0103] Secondly, the risk indices of all valid events in a single trip are summed and divided by the total mileage of the trip to calculate the risk density per unit mileage. .

[0104] By normalizing mileage, the evaluation differences caused by variations in the length of a single trip are eliminated, thus reducing risk density. The formula is as follows: .

[0105] This indicates the mileage of a single trip. This indicates the number of aggressive driving incidents identified.

[0106] Finally, the score for a single trip is calculated using an exponential decay model, as shown in the formula: .

[0107] Represented as the risk penalty coefficient, it determines the gradient characteristics of the exponential decay function. It is configurable and adaptive, and is an empirical constant fitted by the platform based on historical big data statistics, with a selection range between 0.1 and 1.0.

[0108] To meet the intuitive interaction needs of connected vehicle terminals or ride-hailing clients, continuous comfort scores will be used. The comfort level is divided into four discrete levels, and corresponding trip comfort labels are generated.

[0109] If the score is [90, 100], it is marked as imperceptible / extremely comfortable. This indicates that the entire ride was excellent, with no significant adverse events or only very slight energy fluctuations, and the passenger's physiological state remained stable at a baseline throughout the journey.

[0110] If the score is [75, 90), it is marked as a minor level / standard comfort. This indicates that there is an occasional operational irregularity that produces an acceptable range of longitudinal or lateral shocks, causing the passenger to experience slight but quickly subsiding physiological fluctuations.

[0111] If the score is [60, 75), it is marked as noticeable / mild discomfort. This indicates that there is a high density of aggressive events per unit distance, with multiple sudden decelerations or lane changes that can cause group average heart rate variability. Passengers experience turbulence and are prone to motion sickness.

[0112] If the score is [0, 60), it is marked as severe discomfort / severe panic. This situation indicates that a violent physical impact (such as emergency avoidance or reckless driving) has occurred, accompanied by a very high causal confidence level in the model output, and severe abnormalities in the passenger's physiological data.

[0113] By leveraging the nonlinear decay characteristics of the exponential function, this scoring model ensures that the score decreases gradually in the safe range (i.e., low-risk density state), providing drivers with a reasonable margin of error; while in the high-risk range (i.e., high-risk density state), the score drops sharply, thus exerting a strong safety constraint and warning effect on extremely high-risk drivers.

[0114] This invention proposes a multimodal nonlinear fusion algorithm based on dynamic baseline and XGBoost model, which breaks through the limitations of traditional linear threshold evaluation and significantly improves the generalization ability and recognition accuracy of aggressive driving behavior recognition model under different road conditions, different vehicle speeds and different passenger physical conditions.

[0115] Step 5. Utilize cloud server data to record the driver's historical trip information, introduce a time decay factor to conduct a long-term evaluation of the driver, and obtain a comprehensive driver score to assess vehicle ride comfort. Figure 5 As shown.

[0116] In the process of calculating the overall score, a collective intelligence perception mechanism is introduced to correct the sensitivity of passengers.

[0117] This invention focuses not only on the evaluation results of a single trip, but also on the long-term data profiles of drivers and passengers. Therefore, a cloud server is set up to record the driver's historical trip data, and a time decay factor is introduced to conduct a scientific long-term evaluation.

[0118] Meanwhile, a crowd-sensing mechanism is added to the cloud server to correct passenger sensitivity, thereby helping to improve the accuracy of model recognition.

[0119] Traditional evaluation systems assume that all passengers have the same psychological tolerance, which is clearly not true. Some passengers are naturally sensitive (easily startled), and their heart rate spikes at the slightest bump; others are less sensitive or have a habit of driving fast.

[0120] Therefore, this invention establishes a collective intelligence perception mechanism, utilizing the platform's massive historical data to calculate the sensitivity of each passenger relative to the "group average level," thereby adjusting the scoring weights. This invention establishes a closed-loop credit system based on collective intelligence perception for bias correction and time decay, achieving both objective fairness and dynamic evolution of evaluation results. It constructs a dynamic credit profiling system that balances horizontal fairness (eliminating individual passenger bias) and vertical incentives (focusing on recent driver improvements). The process is as follows: I. Global baseline calculation: Let the average physiological stress response value of all passengers on the statistical platform when they experienced the same physical stimulus in the vehicle be set as . .

[0121] ; in The total number of historical samples that experienced the same physical stimulus and were retrieved from the cloud-based historical database. For the first The actual physiological stress response values ​​of passengers in each historical sample.

[0122] II. Individual baseline calculation: Based on the average physiological stress response value of each individual passenger P over their historical multiple trips, it is denoted as: : .

[0123] in The total number of historical valid samples of the target passenger in the cloud database under the same physical stimuli; For this target passenger in history The actual physiological stress response values ​​extracted and recorded in the same physical stimulus.

[0124] The aforementioned equivalent physical stimulus refers to the physical impact energy calculated when the historical event occurred. The physical impact energy of the current aggressive driving incident Within the tolerance range, that is, satisfying: ;in, The energy matching tolerance coefficient is set to control the retrieval tolerance of similar scenarios, and its value range is set to 0.1~0.2.

[0125] III. According to and Two metrics were used to calculate the relative sensitivity of this individual passenger among all passengers on the platform. .

[0126] .

[0127] in This is an adjustment factor used to control the correction strength, and its value ranges from 0.5 to 1.5. The function is used to obtain the relative sensitivity Limit to the set threshold and threshold Between these points, to prevent over-correction.

[0128] and Its function is to serve as a restriction The threshold.

[0129] This is the lower threshold to prevent the weight from being too low due to passengers being overly sensitive; the range is set to 0.3~0.5. It is an upper limit threshold to prevent excessive weighting due to passengers being too slow to react; the range is set to 1.5~2.

[0130] IV. Construct the time decay factor.

[0131] Driving skills and habits are dynamic. A driver who frequently violated traffic rules two years ago but has recently been behaving well should be considered a good driver; conversely, an experienced driver who has recently been committing frequent violations should be given serious warnings.

[0132] Therefore, a time decay factor is introduced to construct a dynamic profile of the driver's driving behavior, rather than a simple arithmetic average.

[0133] Assuming the driver has a history of The next trip, the first The rating for this trip is The time since the occurrence is Then time weight The calculation formula is: .

[0134] in The forgetting rate, used to determine the decay rate of the weights, ranges from 0.001 to 0.1 and can be adjusted according to management needs. This yields the time decay factor. It was later incorporated into the long-term driving credit evaluation of drivers.

[0135] V. Driver profile output and feedback.

[0136] Driver's overall score Calculation formula: .

[0137] Cloud computing acquisition Subsequently, to meet the intuitive interaction needs of vehicle networking terminals or ride-hailing clients, the continuous comfort score is divided into four discrete levels of perceived comfort, and corresponding trip-related perception labels are generated.

[0138] If the score is [90, 100], it is marked as extremely stable. This indicates that: The vehicle's longitudinal or lateral jerk remains at extremely low levels, with little or no triggering of the kinematic threshold. Changes in passenger physiological characteristics ( The ride closely follows the stable baseline, and the collective intelligent perception correction mechanism did not detect any abnormal group discomfort. The riding experience reaches the highest level of comfort, with no risk of motion sickness or fright.

[0139] If the score is [75, 90), it is marked as normal driving. This situation indicates that there are occasional slight acceleration, deceleration or lane changes that generate a small amount of low-intensity physical impact energy. ), but physiological fear energy ( There was no significant jump, and the ride quickly recovered over time, with the riding experience remaining within the normal tolerance threshold for the general public.

[0140] If the score is [60, 75), it is marked as aggressive driving. This indicates that: Frequent and high-density occurrence of effective radical events, response delay time ( The features exhibit a high degree of temporal coupling causality, and are further analyzed by the crowd perception coefficient ( Even after excluding a few passengers with highly sensitive constitutions, the average physiological stress of the group still showed a significant negative shift, easily inducing passengers to experience physiological reactions such as dizziness, anxiety, or mild panic.

[0141] If the score is [0, 60), it is marked as high-risk driving. This indicates that: The confidence probability of a single event involving emergency braking or high-speed cornering. Approaching 1.0, accumulating enormous physical and physiological energy in a short period of time, or having a historically poor credit score with a time decay factor. The effects have not been eliminated, severely compromising passenger comfort and posing high safety risks.

[0142] The final output includes a multi-dimensional driver profile and a comprehensive driver score. This can be used to determine daily order acceptance permissions, setting a trigger threshold based on the driver's overall rating. If the threshold value is below the set threshold, mandatory training will be triggered.

[0143] Through the above steps, this invention establishes a closed-loop evaluation system with adaptive correction capabilities. On the one hand, the collective intelligence perception mechanism effectively solves the problem of misjudgment and omission caused by individual physical differences in single biological feedback evaluation by introducing a sensitivity coefficient by comparing the differences between individuals and the group, thus achieving objectivity and fairness in the evaluation. On the other hand, the time decay model performs weighted cleaning of historical data, giving the evaluation results a distinct timeliness characteristic. It can not only keenly capture the recent trends of drivers' behavior, but also quickly reflect the improvement results of drivers' safe driving, thereby constructing a dynamically evolving driver credit ecosystem.

[0144] This invention can objectively and fairly evaluate the driver's driving behavior based on the passenger's physiological characteristics and the vehicle's motion characteristics, and can create a long-term profile of the passenger's physiological changes when stimulated and the reliability of the driver's driving behavior. It can be used for personalized matching of ride-hailing orders, flexible adjustment of ride-hailing insurance premiums, and improvement of ride-hailing operation safety.

[0145] Example 2 This embodiment 2 describes a vehicle ride comfort evaluation system, which is based on the same inventive concept as the vehicle ride comfort evaluation method in embodiment 1 above.

[0146] A vehicle ride comfort evaluation system includes the following modules: The preprocessing module is used to simultaneously acquire passenger physiological data and vehicle motion data, then perform motion artifact detection and removal, and then perform preliminary extraction on the removed data to obtain passenger physiological data and vehicle motion data. The normalization module identifies the baseline value of passenger physiological data when the vehicle is driving smoothly, and performs Z-Score normalization on the real-time passenger physiological data based on the baseline value to obtain normalized physiological features. The feature extraction module is used to extract feature variables from the standardized passenger physiological data and vehicle motion data; it also extracts two multimodal spatiotemporal coupling features, namely response delay time and physiological-physical energy coupling ratio. The model building and single-trip scoring module is used to build an aggressive driving behavior recognition model. The model input includes extracted vehicle motion features, passenger physiological features, and two multimodal spatiotemporal coupled features. The model output layer uses a Sigmoid function to map and output the confidence probability of aggressive driving behavior in the current time window; further, a dynamic scoring algorithm based on mileage normalization is executed based on the confidence probability to calculate the score for a single trip; And a comprehensive scoring module, which uses cloud server to record the driver's historical trip data, introduces time decay factor to evaluate the driver over a long period of time, and obtains the driver's comprehensive score to evaluate the vehicle's ride comfort. A collective intelligence perception mechanism is introduced into the process of single trip and driver comprehensive scoring to correct passenger sensitivity.

[0147] It should be noted that any content not mentioned in the above-described functional modules of the system described in this embodiment can be referred to the step description of the corresponding method in Embodiment 1 above, and will not be repeated in detail here.

[0148] Example 3 This embodiment 3 describes a computer device including a memory and one or more processors. Executable code is stored in the memory. When the processor executes the executable code, it implements the steps of the vehicle ride comfort evaluation method in embodiment 1 above.

[0149] In this embodiment, the computer device can be any device or apparatus with data processing capabilities, and will not be described in detail here.

[0150] Example 4 This embodiment 4 describes a computer-readable storage medium storing a program that, when executed by a processor, is used to implement the steps of the vehicle ride comfort evaluation method in embodiment 1 above.

[0151] The computer-readable storage medium can be an internal storage unit of any device or apparatus with data processing capabilities, such as a hard disk or memory, or an external storage device of any device with data processing capabilities, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc.

[0152] Of course, the above description is only a preferred embodiment of the present invention. The present invention is not limited to the above-described embodiments. It should be noted that any equivalent substitutions or obvious modifications made by those skilled in the art under the guidance of this specification fall within the scope of this specification and should be protected by the present invention.

Claims

1. A method for evaluating vehicle ride comfort, characterized in that, Includes the following steps: Step 1. Simultaneously acquire passenger physiological data and vehicle motion data, then perform motion artifact detection and removal, and then perform preliminary extraction on the removed data to obtain passenger physiological data and vehicle motion data; Step 2. Identify the baseline values ​​of passenger physiological data when the vehicle is in a stable driving state, and perform Z-Score standardization on the real-time collected passenger physiological data based on the baseline values ​​to obtain normalized physiological features; Step 3. Extract feature variables from the standardized passenger physiological data and vehicle motion data; simultaneously extract two multimodal spatiotemporal coupling features, namely response delay time and physiological-physical energy coupling ratio; Step 4. Build an aggressive driving behavior recognition model, wherein the input of the model includes the vehicle motion features, passenger physiological features and two multimodal spatiotemporal coupling features extracted in Step 3; The model output layer uses a Sigmoid function to map and output the confidence probability of aggressive driving behavior in the current time window; further, a dynamic scoring algorithm based on mileage normalization is executed based on the confidence probability to calculate the score for a single trip; Step 5. Use the cloud server to record the driver's historical trip data, introduce a time decay factor to conduct long-term evaluation of the driver, and obtain the driver's comprehensive score to evaluate the vehicle ride comfort. A collective intelligence perception mechanism is introduced into the process of single trip and driver comprehensive scoring to correct passenger sensitivity.

2. The vehicle ride comfort evaluation method according to claim 1, characterized in that, Step 1 specifically involves: Step 1.

1. Extract passenger physiological data, including HRV, using wearable devices, and simultaneously acquire vehicle motion data using in-vehicle inertial navigation and GPS. Step 1.

2. Based on the wrist inertial sensor data continuously collected by the wearable device, calculate the intensity of motion within a local time window in order to identify and eliminate motion artifacts of physiological signals caused by the passenger's own limb movements; Step 1.

3. Perform preliminary extraction on the discriminated data to obtain the physiological data, including HRV, and vehicle motion data, including longitudinal velocity, acceleration, jerk, and vehicle position information, required for subsequent model input.

3. The vehicle ride comfort evaluation method according to claim 2, characterized in that, Step 1.2 specifically includes: I. Calculation of wrist composite acceleration; Collect three-axis acceleration data from wearable devices , , ; Calculate the magnitude of the combined wrist acceleration: ； in This represents the instantaneous composite acceleration of the wrist at time t; II. Dynamic signal separation; A robust sliding window variance method is used, and the window size is set. The formula for calculating the standard deviation of wrist acceleration is as follows: ； in It is the mean within the window. This represents the standard deviation of acceleration measured within the sliding time window; III. Motion artifact detection; Set motion interference threshold Motion artifact mask formula: ； in Indicates the signal confidence mask; when When the confidence score is 0, the current physiological signal is determined to be contaminated by motion noise and is discarded directly, or its corresponding confidence weight is set to 0; when... If = 1, then retain it.

4. The vehicle ride comfort evaluation method according to claim 2, characterized in that, Step 2 specifically involves: Step 2.

1. Monitor vehicle status using the collected vehicle motion data and GPS data, and define a time period that meets the following conditions as a stable window: longitudinal acceleration accelerometer Duration ; Step 2.

2. Within the stable window, calculate the moving average and standard deviation of the passenger HRV index; Step 2.

3. Perform real-time standardization using Z-score. Use the moving average and standard deviation of passenger HRV index calculated within the stationary window to transform the real-time HRV values ​​at subsequent time points.

5. The vehicle ride comfort evaluation method according to claim 1, characterized in that, In step 3, the extraction process for response delay time and physiological-physical energy coupling ratio is as follows: I. Response latency , used to describe the time difference between the occurrence of a physical stimulus and the peak time of a physiological response; ； in The value is the Z-Score standardized value of the HRV calculated in real time in step 2. Extract the calculated jerk from step 1; This indicates the time corresponding to when the maximum value is reached; The physiological response of the human body to fright is usually delayed by 0.5s to 3s; if the response delay τ is within this range, it is judged as a real radical event; if τ < 0, that is, the reaction occurs before braking, or τ > 10s, it is judged as noise. II. Physiological-physical energy coupling ratio This is used to measure whether the "stimulus" and "response" match, and the formula is as follows: ； in To prevent tiny constants with a denominator of zero, and Both are scalar values; based on The sequence is used to integral the square of the passenger's stress level within the response time window after the physical event occurs, while also introducing... The function, which is the squared energy integral of the positive deviation representing the physiological stress state, is given by the following formula: ； based on Sequence, setting aggressive trigger threshold ,when > When a suspected aggressive driving incident is identified, a time window of the incident is captured, and the square integral of the acceleration within that time window is performed, as shown in the following formula: ； for and Time window for physical energy calculation It is based on the absolute value of the instantaneous acceleration first exceeding the radical trigger threshold. Open the time window at that time; Until Falling back to the safe exit threshold And the time window is closed after maintaining the preset duration of 1 second; The moment when the event begins, i.e. The time is identified as the start time of the event. Start energy integration; The end time of the event, i.e. when When, mark as the end time of the event. Stop integrating.

6. The vehicle ride comfort evaluation method according to claim 1, characterized in that, In step 4, after extracting feature variables from passenger physiological data and vehicle motion data, a total of 24 HRV features and 10 vehicle motion features were extracted.

7. The vehicle ride comfort evaluation method according to claim 1, characterized in that, In step 4, the process of calculating the score for a single trip using a dynamic scoring algorithm based on mileage normalization is as follows: The Sigmoid function is used to map the raw XGBoost predictions to aggressive driving confidence probabilities. After obtaining the confidence scores of each aggressive driving event identified by the model, dynamic scoring based on risk density is performed. First, calculate the risk index for a single effective aggressive event. , Defined as passenger sensitivity coefficient recorded in the cloud Confidence probability of model output With the physical energy of the event The product of the logarithmic functions is calculated using the following formula: ； Secondly, the risk indices of all valid events in a single trip are summed and divided by the total mileage of the trip to calculate the risk density per unit mileage. Risk density The formula is as follows: ； in, Indicates the mileage of a single trip. This indicates the number of aggressive driving incidents identified; Finally, the score for a single trip is calculated using an exponential decay model. The formula is: ； in This is expressed as a risk penalty coefficient; Score for a single trip The comfort level is divided into four discrete levels, and corresponding trip comfort labels are generated.

8. The vehicle ride comfort evaluation method according to claim 1, characterized in that, In step 5, the formula for calculating the driver's overall score is as follows: I. Global benchmark calculation: Let the average physiological stress response value of all passengers on the statistical platform when they experienced the same physical stimulus in the vehicle be set as . : ； in The total number of historical samples that experienced the same physical stimulus and were retrieved from the cloud-based historical database; For the first The actual physiological stress response values ​​of passengers in each historical sample; II. Individual baseline calculation: Based on the average physiological stress response value of each individual passenger P over their historical multiple trips, it is denoted as: : ； in The total number of historical valid samples of target passengers under the same physical stimuli in the cloud database; For target passengers in history The actual physiological stress response values ​​extracted and recorded in the same physical stimulus; Equal physical stimulus refers to the physical impact energy calculated when the historical event occurred. The physical impact energy of the current aggressive driving incident Within the tolerance range; III. According to and Two metrics were used to calculate the relative sensitivity of an individual passenger among all passengers on the platform. ; ； in This is an adjustment factor used to control the correction strength, and its value ranges from 0.5 to 1.

5. The function is used to obtain the relative sensitivity Limit to the set threshold and threshold Between these points, to prevent over-correction; and Its function is to serve as a restriction The threshold; This is the lower threshold to prevent the weight from being too low due to passengers being overly sensitive; the range is set to 0.3~0.

5. This is the upper limit threshold to prevent excessive weighting due to passengers being too slow to react; the range is set to 1.5~2. IV. Construct a time decay factor to create a dynamic profile of the driver's driving behavior; assume the driver has a history of... The next trip, the first The rating for this trip is The time since the occurrence is Then time weight The calculation formula is: ； in The forgetting rate is used to determine the rate at which the weights decay. V. Defining the Driver's Overall Score The calculation formula is as follows: ； The final output includes a multi-dimensional driver profile and a comprehensive driver score. ; The driver's final comprehensive score is calculated in the cloud. Then, the continuous comfort score is divided into four discrete levels of perceived comfort, and corresponding trip-related perception labels are generated.

9. The vehicle ride comfort evaluation method according to claim 8, characterized in that, Determine the physical impact energy of the current aggressive driving incident. The formula for the tolerance range is as follows: ； in The set energy matching tolerance coefficient has a value range of 0.1 to 0.

2.

10. A vehicle ride comfort evaluation system, characterized in that, Includes the following modules: The preprocessing module is used to simultaneously acquire passenger physiological data and vehicle motion data, then perform motion artifact detection and removal, and then perform preliminary extraction on the removed data to obtain passenger physiological data and vehicle motion data. The normalization module identifies the baseline value of passenger physiological data when the vehicle is driving smoothly, and performs Z-Score normalization on the real-time passenger physiological data based on the baseline value to obtain normalized physiological features. The feature extraction module is used to extract feature variables from the standardized passenger physiological data and vehicle motion data; it also extracts two multimodal spatiotemporal coupling features, namely response delay time and physiological-physical energy coupling ratio. The model building and single-trip scoring module is used to build an aggressive driving behavior recognition model. The model input includes extracted vehicle motion features, passenger physiological features, and two multimodal spatiotemporal coupled features. The model output layer uses a Sigmoid function to map and output the confidence probability of aggressive driving behavior in the current time window; further, a dynamic scoring algorithm based on mileage normalization is executed based on the confidence probability to calculate the score for a single trip; And a comprehensive scoring module, which uses cloud server to record the driver's historical trip data, introduces time decay factor to evaluate the driver over a long period of time, and obtains the driver's comprehensive score to evaluate the vehicle's ride comfort. A collective intelligence perception mechanism is introduced into the process of single trip and driver comprehensive scoring to correct passenger sensitivity.

Images