Brain-controlled car system performance evaluation and evaluation method for key test scenarios

Abstract

The application faces the brain-controlled car system performance evaluation and evaluation method of key test scene, belongs to the cross field of intelligent automobile and brain-computer interaction technology. This method constructs a four-level risk model based on the "environment-task-interference" three-dimensional factor, and achieves a more scientific risk division. The model can cover complex scenes such as extreme weather and emergency tasks, and can accurately identify the performance defects of the system under high-risk working conditions, and clearly direct the system optimization. Based on the newly added "data security" dimension, a more comprehensive evaluation system is constructed, which makes up for the lack of brain electrical privacy data protection in the existing evaluation, meets the national data compliance requirements, and effectively enhances the integrity and compliance of the evaluation system. Based on 5G+TSN technology, nanosecond-level data synchronization is realized, a virtual-real closed-loop test platform is created, the simulation scene and the dynamics characteristics of the real car are highly consistent, the evaluation result is more close to the engineering actual application, and the reliability and practicality of the evaluation are greatly improved.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of intelligent vehicles and brain-computer interface technology, and in particular relates to the performance evaluation and assessment method of brain-controlled vehicle system for key testing scenarios. Background Technology

[0002] Brain-controlled vehicle (BCV) systems achieve contactless control of vehicle functions by decoding the driver's brainwave signals, representing an important development direction for human-machine interaction in intelligent vehicles. However, existing BCV system testing technologies still suffer from the following core technical problems, making it difficult to meet the requirements of engineering applications: The risk classification of test scenarios is crude: the current scenario library is mostly divided into "high / low risk" without considering environmental complexity, such as extreme weather such as fog and rainstorms, the urgency of driving tasks such as normal following and emergency obstacle avoidance, the intensity of EEG signal interference, such as the coupling mechanism of single visual stimulation and multimodal composite stimulation. This classification method results in low coverage of the real driving environment, which cannot accurately expose the performance defects of the system under extreme conditions and is difficult to support safety verification in high-risk scenarios.

[0003] The performance evaluation system is incomplete: existing evaluation indicators focus on traditional dimensions such as driving safety and timeliness, lacking consideration for "data security." Specifically, brain-controlled vehicle systems involve the driver's brainwave data, which falls under the category of sensitive biometric information. Current assessments do not cover key security indicators such as data encryption strength, packet loss rate, and privacy anonymization rate, failing to meet the protection requirements for sensitive biometric data under laws and regulations such as the Personal Information Protection Law, thus posing data compliance risks.

[0004] The testing platform lacks a closed-loop virtual-real environment: most evaluation solutions rely on pure simulation environments or simple real-vehicle tests, resulting in a significant disconnect between the two. Simulation scenarios cannot accurately replicate the dynamic characteristics of real vehicles, and the synchronization accuracy of EEG data, vehicle data, and scene data can only reach the millisecond level, making it difficult to truly reproduce the dynamic interaction process between the brain, vehicle, and environment. This deficiency leads to discrepancies between the evaluation results and the actual needs of engineering applications, failing to provide accurate basis for system optimization.

[0005] To address the aforementioned technical issues, there is an urgent need to construct an integrated evaluation scheme that includes "scientific risk classification, comprehensive indicator evaluation, virtual and real closed-loop testing, and standard experimental implementation" to provide technical support for the performance optimization and engineering implementation of the brain-controlled vehicle system. Summary of the Invention

[0006] This invention aims to address the technical pain points in the evaluation of existing brain-controlled vehicle systems, such as "unscientific scenario risk classification, lack of data security dimension in the evaluation system, lack of virtual-real closed loop in the testing platform, and non-reproducible experimental schemes," and provides a quantifiable, reproducible, and comprehensive performance evaluation and assessment method.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for performance evaluation and assessment of brain-controlled vehicle systems for key testing scenarios, comprising the following steps: Step S1: Constructing the test scenario; By integrating a driving simulator and an EEG device, a cloud platform for synchronizing human-vehicle-environment information is constructed to simulate diverse road environments, driving tasks, and in-vehicle environmental interference scenarios. The platform can collect EEG signals, vehicle motion parameters, and driving scenario data in real time through the simulator and EEG device, achieving synchronous acquisition of multi-source data. At the same time, the Savitsky-Golay filtering algorithm is used to denoise the vehicle motion parameters. Step S2: Construct a three-dimensional risk factor model of environmental complexity, task urgency, and interference intensity; Based on the data from step S1, a three-dimensional risk factor model of environmental complexity, task urgency, and interference intensity is constructed. A key test scenario library with four risk levels is also constructed based on the risk factors. The risk factors for environmental complexity include weather type and traffic flow. The risk factors for task urgency include driving task type, minimum headway (THW), and minimum time to collision (TTC). The risk factors for interference intensity include stimulus mode and interference duration. The four risk levels are extremely low risk, low risk, medium risk, and high risk, respectively. Step S3: Setting up key test scenarios; The orthogonal experimental design method is used to determine the combination scheme of each risk factor in step S2, which needs to cover all four levels of risk scenarios. Specifically, the risk factors of environmental complexity, task urgency and interference intensity are orthogonally combined first. Then, the risk factor scenarios after orthogonal combination are selected, and four basic scenarios are selected according to risk weight. Finally, the basic scenarios of environmental complexity, task urgency and interference intensity are orthogonally combined again to form 4×4×4=64 basic scenarios. Step S4: Virtual simulation modeling; A joint simulation platform is used to construct virtual simulation scenarios and high-fidelity dynamic models. Scenarios involving risk factors are selectively added by the orthogonal experiments in step S3. Step S5: Real-world scenario reproduction; By modifying the actual vehicle site, laying a controllable water / snowmaking system and setting dynamic obstacles, a one-to-one mapping between the simulation scene and the actual vehicle scene can be achieved. Step S6: Multi-dimensional performance evaluation indicator system; A multi-dimensional performance evaluation index system with a three-layer structure of target layer, overall index layer, and index layer is constructed: the target layer is for multi-dimensional performance evaluation of the brain-controlled vehicle system, the overall index layer includes six dimensions: driving safety, anti-interference, generalization, timeliness, comfort, and data security, and the index layer consists of specific quantitative evaluation indicators under each overall index. The evaluation indicators for driving safety are TTC, THW, lane centerline deviation (LCO), and emergency braking distance (EBD); the evaluation indicators for anti-interference are function recognition accuracy (ERA), false trigger rate (FPR), and signal-to-noise ratio (SNR); the evaluation indicators for generalization are cross-subject accuracy (CPA) and cross-scenario adaptation rate (CSA); the evaluation indicators for timeliness are command response time (RT), information transmission rate (ITR), and command execution delay (ED); the evaluation indicators for comfort are mental workload (ML), device wearing comfort (WC), and operation satisfaction (OS); and the evaluation indicators for data security are data encryption strength (ES), data transmission packet loss rate (PLR), privacy data anonymization rate (PDR), and data storage integrity (ESI). The calculation method for TTC, one of the evaluation indicators for driving safety, is as follows: ; In the formula, For relative distance, The absolute value is taken when the relative velocity is negative. The absolute speed of a vehicle in a continuous series of vehicles; The speed of the vehicle targeting the obstacle; The calculation method is as follows: ; In the formula, The distance between the front ends of vehicles traveling in a continuous line; The calculation method for EBD is as follows: ; In the formula, For system response time, For braking and deceleration, This is the initial braking speed; The LCO is calculated as follows: ; In the formula, The lateral coordinates of the vehicle's center of gravity, acquired via an on-vehicle IMU or LiDAR, are shown perpendicular to the direction of travel. (Unit: [unit not specified]) ; The lateral coordinates of the lane centerline obtained through SCANeR scene modeling or high-precision real-vehicle maps, in units of ; Step S7: Construction of a closed-loop virtual-real testing platform integrating virtual simulation, real vehicle testing, data synchronization, and data security; The virtual simulation platform generates four levels of risk test scenarios: extremely low risk, low risk, medium risk, and high risk. It outputs scenario parameters and virtual vehicle status, and synchronizes scenario commands to the real vehicle subsystem. The subsystem consists of hardware and software systems. The real vehicle testing system is mounted on a modified experimental vehicle. The original driving control module of the experimental vehicle has been removed, and a brain control interface module has been added, which is compatible with the CANFD bus. The real vehicle testing subsystem specifically includes an EEG acquisition module, a vehicle status acquisition module, and a stimulus generation module. The data synchronization system can achieve time alignment of multi-source data from virtual simulation, real vehicles, and EEG, specifically including synchronization protocols, synchronization nodes, and unified data format processing. Data security is deeply integrated with virtual simulation, real vehicle testing, and data synchronization subsystems to protect sensitive biometric information. Step S8: Establish standardized evaluation methods; In step S7, on the four-in-one virtual-real closed-loop testing platform, a multi-dimensional performance evaluation index system is used as the evaluation standard to select suitable experimental groups for training, construct experimental scenarios, execute standardized experiments, acquire and analyze relevant data, and finally complete the establishment of standardized evaluation methods.

[0008] Furthermore, the specific implementation of the Savitsky-Golay filtering algorithm in step S1 is as follows: ; In the formula, To smooth and denoise the data, Here are the original data, where 'a' is the filter radius and 2a+1 is the sliding port length. This is the weighting factor.

[0009] Furthermore, in step S2, the risk factor classification of environmental complexity based on weather type is defined as follows: clear weather with visibility ≥ 20km is defined as extremely low risk; cloudy weather with visibility 10km-20km is defined as low risk; rainy weather with visibility 5km-10km is defined as medium risk; and foggy and snowy weather with visibility ≤ 5km is defined as high risk. The risk level classification based on traffic flow is defined as follows: scenarios with traffic flow ≤ 10 vehicles / km are defined as extremely low risk; scenarios with traffic flow 10-20 vehicles / km are defined as low risk; scenarios with traffic flow 20-30 vehicles / km are defined as medium risk; and scenarios with traffic flow > 30 vehicles / km are defined as high risk. The risk factors for task urgency are categorized as follows: Driving tasks are classified as follows: straight-line constant speed driving is classified as extremely low risk; smooth turning or following other vehicles is classified as low risk; lane changing and obstacle avoidance are classified as medium risk; and emergency braking and collision warning are classified as high risk. The risk levels for time-to-weight (THW) are defined as follows: THW ≥ 5s is classified as extremely low risk; THW 3s-5s is classified as low risk; THW 2s-3s is classified as medium risk; and THW ≤ 2s is classified as high risk. The risk levels for time-to-control (TTC) are defined as follows: TTC ≥ 8s is classified as extremely low risk; TTC 5s-8s is classified as low risk; TTC 3s-5s is classified as medium risk; and TTC ≤ 3s is classified as high risk. The risk level classification of interference intensity is defined as follows: no-stimulus scenarios are defined as extremely low-risk, single-visual-stimulus scenarios are defined as low-risk, visual-auditory composite stimulation scenarios are defined as medium-risk, and visual-auditory-tactile multimodal composite stimulation scenarios are defined as high-risk. The risk level classification of interference duration is defined as follows: no-interference scenarios are defined as extremely low-risk, interference duration of 1s-3s is defined as low-risk, interference duration of 3s-5s is defined as medium-risk, and interference duration of ≥5s is defined as high-risk.

[0010] Furthermore, the 64 basic scenarios in step S3 are arranged in reverse order of risk level, and the top 20 scenarios are selected as key core scenarios.

[0011] Furthermore, in step S4, the virtual simulation scenario is a simulated urban road with six lanes in both directions, each lane being 3.5 meters wide, with double yellow lines in the middle and green belts on both sides, all of which are non-crossable areas. At the same time, static environmental elements are added to ensure the realism of the scenario.

[0012] Furthermore, the threshold values ​​for the driving safety evaluation indicators in step S6 are as follows: The security threshold for TTC is defined as follows: A collision will occur if TTC < 0.5s, and a collision warning should be triggered if TTC < 2.9s. Therefore, the TTC safety threshold is set to ≥ 2.9s. The security threshold of THW is defined as follows: A collision warning needs to be triggered when THW < 2.5s, therefore the THW safety threshold is set to ≥ 2.5s; The safety threshold for EBD is defined as follows: Under the urban road commuting condition with a vehicle speed ≤ 50 km / h, the EBD safety threshold ≤ 30 m; under the condition of an ice and snow road surface with an adhesion coefficient 1 / 3 - 1 / 2 that of a dry asphalt road surface, the EBD safety threshold ≤ 50 m; under the highway condition with a vehicle speed ≤ 120 km / h, the EBD safety threshold ≤ 120 m; The safety threshold of LCO is defined as: On a structured road, LCO ≤ 0.3 m represents a low risk; 0.3 m < LCO ≤ 0.8 m represents a medium risk; LCO > 0.8 m represents a high risk; The specific calculation methods of the evaluation indexes of anti-interference, generalization, timeliness, comfort, and data security are as follows: Anti-interference: The calculation method is as follows, ; In the formula, is the number of correct identifications, is the number of incorrect identifications; Definition: For safety-critical functions, the ERA safety threshold ≥ 95%; for non-safety-critical functions, the ERA safety threshold ≥ 90%; The calculation method of FPR is as follows: ; In the formula, is the number of false positives, is the number of true negatives; Definition: Under the normal driving scenario on a highway on a sunny day, the FPR safety threshold ≤ 0.1%; under the urban road scenario, the FPR safety threshold is relaxed to ≤ 0.5%; The calculation method of SNR is as follows: ; In the formula, represents the effective value of the signal amplitude, represents the effective value of the signal noise amplitude, is the amplitude of the electroencephalogram alpha wave, is the amplitude of the electromyogram interference; Definition: Under the heavy rain condition during daytime on a highway, the SNR safety threshold ≥ 15 dB; under the ordinary urban road condition, the SNR safety threshold ≥ 10 dB; Generalization: The calculation method of CPA is as follows, ; In the formula, is the number of test subjects, is the accuracy rate of the Definition: Basic pass threshold CPA ≥ 80%; Excellent threshold CPA ≥ 85%, which means meeting the multi-user scenario requirements of L2+ level assisted driving system; The calculation method for CSA is as follows: ; In the formula, For the number of scenes, For the scene Number of times the function was successfully implemented in China This represents the total number of tests. Definitions: Low-difference scenario, from sunny to cloudy urban areas, CSA ≥ 90%; Medium-difference scenario, from sunny to rainy urban areas, and from urban areas with streetlights at night, CSA ≥ 80% for rainy days and CSA ≥ 85% for nighttime; High-difference scenario, from sunny to rainy urban areas without streetlights, CSA ≥ 70%. Timeliness: The calculation method for RT is as follows, ; In the formula, To record the timing of EEG intention triggering based on the TSN time synchronization system, This refers to the timing of command execution by the vehicle ECU based on the TSN time synchronization system. Definition: In high-speed cruising scenarios, the normal response time (RT) is ≤0.5s, and in emergency scenarios, the RT is ≤0.3s; in urban road scenarios, the RT is ≤0.8s. The calculation method for ITR is as follows: ; In the formula, For single selection of time, For the number of optional targets, For accuracy; Definition: In the context of highways, the basic ITR is ≥8 bits / s, the total transmission time of critical information is ≤0.3s, and the proportion of redundant information is ≤10%. The calculation method for ED is as follows: ; In the formula, Record the CANFD bus command transmission time. The moment when the vehicle actuator operates; Definitions: Under normal high-speed operation, ED ≤ 0.5s; under high-speed emergency avoidance, ED ≤ 0.3s; under low-speed urban scenarios, ED ≤ 1s. Comfort: The calculation method for ML is as follows, ; In the formula, For dimension weights, The dimension score (0-100 points) is where n represents the core dimension of ML. The threshold for mental workload (ML) needs to be set in conjunction with the driving scenario. In a normal driving scenario, ML ≤ 3 points is considered too low load, corresponding to a driving state of inattention and slow reaction. ML 3-7 points is considered a comfortable load, corresponding to a driving state of focused attention and easy cognition. ML > 7 points is considered too high load, corresponding to a driving state of cognitive fatigue and decision-making errors. The NASA-TLX scale is used to calculate the weighted average score of six dimensions, including mental workload, physical workload, time workload, effort level, frustration, and performance. The calculation method for WC is as follows: ; In the formula: The overall score for the comfort of wearing the device is given, ranging from 0 to 1. The closer the score is to 1, the higher the comfort level. A 9-point comfort scale was used to rate driver comfort. This is an objective physiological indicator, expressed as local pressure, in kPa. This is an objective physiological indicator, expressed as changes in skin moisture, with units of g / m². It is an objective physiological indicator, expressed as muscle tension, with the unit being μV; The OS is calculated as follows: ; In the formula, OS is the total operation satisfaction score, which ranges from 0 to 100. The weight of the i-th evaluation dimension is set according to the importance of the scenario, and the sum of the weights of all dimensions is 1. The score for the i-th evaluation dimension ranges from 0 to 10; n is the number of evaluation dimensions. Data security: The calculation method is as follows: ; In the formula, The final data encryption strength score has a range of values. A score of 80 or above is considered Scene safety compliance standards; , , Weight coefficients for each dimension, satisfying The weights need to be based on Specific sub-scenes are dynamically adjusted. The encryption algorithm is rated on its anti-cracking complexity [0, 100] to measure its ability to resist brute-force and differential attacks; The security score for the entire lifecycle management of keys is [0, 100], which measures the protection capabilities of key generation, storage, update, and destruction. The adaptability score for data transmission / storage scenarios is [0, 100]. In urban road vehicle-to-vehicle communication (V2V) conditions, the security threshold for ES (Electronic Security) is ≥7 points. The specific classifications and corresponding risks are as follows: An ES score of 8-10 points indicates a high security level, with extremely strong resistance to hacking and the ability to withstand national-level computing power attacks. It ensures data authenticity and integrity, and there are no security risks in system decision-making. An ES score of 7-8 points represents the security threshold, capable of resisting conventional hacker attacks, ensuring data transmission and storage security; only extreme computing power attacks could lead to risks. An ES score of 5-7 points indicates a low security level, easily cracked by professional hackers, potentially leading to data tampering and increased risk of collision warning delays. An ES score <5 points indicates an insecure level, with no effective encryption or easily cracked encryption, making data highly susceptible to tampering. This could cause the system to misjudge distances, miss collision warnings, and lead to accidents. The calculation method for PLR is as follows: ; In the formula, For the number of lost packets, Total number of data packets; In urban V2X operation, with high vehicle density, frequent lane changes, and numerous unexpected scenarios, a PLR (Probability of Collision) of 0-0.5% represents an extremely safe level, with almost no data loss, a calculation error of <1%, and a warning response delay of <10ms, fully meeting safety decision-making requirements. A PLR of 0.5%-2% represents a safe level, where occasional packet loss can be compensated for by data retransmission mechanisms, with a calculation error of <3% and a warning delay of <50ms, posing no safety risk. A PLR of 2%-5% represents a critical risk level, where packet loss frequency increases and retransmission mechanisms cannot fully cover the impact, resulting in a calculation error of 5%-8% and a warning delay exceeding 100ms, requiring a speed reduction warning. A PLR >5% represents a high-risk level, where significant data loss leads to calculation distortion and warning failure, directly increasing the risk of collision and necessitating the triggering of an emergency avoidance mode. The calculation method for PDR is as follows: ; In the formula, This refers to the amount of sensitive data that has been anonymized. Total amount of sensitive data; Intelligent driving on urban roads is a scenario with high privacy risks, requiring a PDR (Performance Recognition Disclosure) rate of ≥95%, with a 100% anonymization rate for key features involved in identity recognition. The calculation method for DSI is as follows: ; In the formula, Total amount of data stored To determine the total amount of data to be stored, the DSI value range is: ;when When the value is 1, it means that all stored data has passed the integrity check and there is no damage; when the value is 1, it means that all stored data has passed the integrity check and there is no damage. When this occurs, it means that all stored data is corrupted or lost, resulting in a complete loss of integrity. Definition: When DSI ≥ 99.9%, it is a safe level, the core data is extremely complete, the calculation is accurate, and there is no safety risk; when DSI is in the range of 99.0% to 99.9%, it is a warning level, a small amount of non-core data is missing or damaged, resulting in a slight deviation in calculation accuracy; when DSI < 99.0%, it is a dangerous level, the core data is lost or tampered with, and it is necessary to urgently downgrade to manual driving.

[0013] Furthermore, the virtual simulation platform in step S7 adopts the SCANeR Studio 2025 and Prescan co-simulation platform. The hardware system of its subsystem specifically includes an industrial-grade server, an Intel Xeon Gold 6426R CPU, an NVIDIA A100 GPU, 256GB DDR5 memory, and a 4K three-screen projection system; the software system specifically includes the scene modeling software SCANeR Studio 2025, the vehicle dynamics simulation software Prescan 2024, and the EEG signal simulation software MATLAB / Simulink. The EEG acquisition module of the real vehicle test system is equipped with a Neuroscan SynAmps RT multi-channel EEG instrument: 64 channels, 1000Hz sampling rate, input impedance ≥10GΩ, and Ag / AgCl electrode device to collect information from key brain regions, with specific sampling points covering key brain regions. The vehicle status acquisition module is equipped with millimeter-wave radar: detection range 0.1-200m, accuracy ±0.1m; lidar: 128 lines, point cloud density 200 points / ㎡; inertial measurement unit (IMU): sampling rate 1000Hz, acceleration accuracy ±0.01m / s². The stimulation generation module is equipped with an in-vehicle display screen for visual cues, an in-vehicle audio system for auditory cues (frequency 500-1000Hz), and a seat vibration module for tactile cues (vibration frequency 20-50Hz, amplitude 0.5-2mm). The data synchronization system uses a 5G+Time Sensitive Network (TSN) synchronization protocol. The synchronization node is achieved by deploying three time synchronization nodes: a scenario node for the simulation subsystem, an EEG node for the EEG acquisition module, and a vehicle node for the real vehicle subsystem. Each node is connected through a Time-Sensitive Network (TSN) switch. The data is uniformly encapsulated in JSON-LD format, including timestamp UTC time and local nanosecond offset; the data types specifically include scene, EEG, and vehicle; the data content includes EEG amplitude and vehicle speed.

[0014] Furthermore, the preparation stage of the standardized evaluation method in step S8 is as follows: Selecting and training suitable candidates: Candidates must meet the following screening criteria: age 25-40, driving experience ≥3 years and holding a C1 or above driver's license, no brain or mental illness, normal vision or corrected vision ≥1.0, pass the Mini-Mental State Examination (MMSE) with a score ≥27, which indicates normal cognitive ability. The training includes a 2-hour pre-experiment session, specifically covering training on wearing the EEG device to ensure electrode impedance ≤5kΩ to guarantee the accuracy of EEG signal acquisition, brain control function training to enable subjects to operate it proficiently, and scenario adaptation training to familiarize subjects with the operational logic of scenarios with different risk levels. After the training, subjects should achieve a brain control function recognition accuracy rate of ≥80%. Experimental scenarios were constructed: Based on the risk level of the scenarios, the experiment was divided into four groups: extremely low risk group, low risk group, medium risk group, and high risk group. Each group contained five core scenarios. Each scenario tested five brain-controlled functions: raising and lowering car windows, adjusting air conditioning temperature, switching music, adjusting volume, and braking. Regarding speed control, the extremely low risk group is 15-30 km / h, the low risk group is 10-25 km / h, the medium risk group is 8-20 km / h, and the high risk group is 5-15 km / h; the stimulation type is divided into weak stimulation and strong stimulation. Each scenario and function combination was tested 10 times to ensure that the experimental data were statistically significant and that the sample size was ≥50.

[0015] Furthermore, the standardized experiment for the standardized evaluation method in step S8 is as follows: Step S81: After the subject arrives at the laboratory, he / she first fills out an informed consent form, then puts on the EEG device, and at the same time performs electrode impedance calibration to ensure that the impedance of all channels is ≤5kΩ to ensure normal acquisition of EEG signals. Step S82: Start the test platform, synchronize the virtual simulation subsystem, the real vehicle subsystem and the data synchronization subsystem, and confirm that the communication between each module is normal; Step S83: Test in order of scenario risk level from low to high to avoid the test results of high-risk scenarios being affected by subject fatigue. Before each scenario begins, the simulation subsystem will output scenario prompts to let the subject know the scenario information. Step S84: During the scenario operation, the subject performs corresponding brain control functions according to the stimulus prompts. During this process, the data synchronization subsystem collects EEG data, vehicle data, and scenario data in real time.

[0016] Furthermore, the data processing method for the standardized evaluation method in step S8 is as follows: Data processing consists of two parts: data preprocessing and data backup. In data preprocessing, the Savitsky-Golay filter with a window length of 5 and a polynomial order of 2 is used to remove jitter from vehicle data. For EEG data, an 8Hz high-pass filter + a 30Hz low-pass filter is used to remove 50Hz power frequency interference and electromyographic noise. In the data backup phase, experimental data is stored in real time on edge computing nodes equipped with encrypted hard drives, and uploaded to the cloud distributed storage Alibaba Cloud OSS within 24 hours. At the same time, a SHA-256 hash value is generated for data integrity verification to prevent data loss or tampering.

[0017] Through the above design scheme, the present invention can bring the following beneficial effects: This invention constructs a four-level risk model based on the three-dimensional factors of "environment-task-interference," achieving a more scientific risk classification. This model can cover complex scenarios such as extreme weather and emergency tasks, accurately identifying system performance defects under high-risk conditions and pointing the way for system optimization. Based on the added dimension of "data security," a more comprehensive evaluation system is formed, filling the gap in existing evaluations regarding the protection of EEG privacy data, complying with national data compliance requirements, and effectively improving the completeness and compliance of the evaluation system. Based on 5G+TSN technology, nanosecond-level data synchronization is achieved, creating a virtual-real closed-loop testing platform that highly matches the simulation scenario with the dynamic characteristics of the actual vehicle, making the evaluation results closer to actual engineering applications and significantly improving the reliability and practicality of the evaluation. Based on clearly defined subject selection criteria, variable control methods, and experimental procedures, the experimental scheme is standardized, ensuring that experiments can be repeated in different laboratories, providing a unified benchmark for performance comparison of brain-controlled vehicle systems, and effectively promoting technical exchange and development within the industry. Attached Figure Description

[0018] The present invention will be further described below with reference to the accompanying drawings and specific embodiments: Figure 1 This is a diagram showing the overall architecture of the virtual-real closed-loop testing platform in an embodiment of the present invention. Figure 2 This is a risk classification model diagram for key test scenarios in an embodiment of the present invention; Figure 3 This is a diagram illustrating the system architecture of the multi-dimensional performance evaluation indicator in this embodiment of the invention. Figure 4 This is a standardized experimental flowchart in an embodiment of the present invention. Detailed Implementation

[0019] The present invention will be further illustrated below with reference to specific embodiments, but the embodiments do not limit the present invention in any way.

[0020] The present invention provides a method for performance evaluation and assessment of a brain-controlled vehicle system for key testing scenarios, comprising the following steps: Step S1: Constructing the test scenario; By integrating the SimEASY driving simulator and an EEG device, a cloud platform for synchronized human-vehicle-environment information is constructed to simulate diverse road environments, driving tasks, and in-vehicle environmental interference scenarios. This platform can acquire EEG signals, vehicle motion parameters, and driving scenario data in real time through the simulator and EEG device, achieving simultaneous acquisition of multi-source data. Simultaneously, the Savitsky-Golay filtering algorithm is used to denoise the vehicle motion parameters, effectively solving the problems of jitter and outlier spikes in the original vehicle data. The specific implementation of the Savitsky-Golay filtering algorithm is as follows: ; In the formula, To smooth and denoise the data, Here are the original data, where 'a' is the filter radius and 2a+1 is the sliding port length. This is the weighting factor.

[0021] SimEASY, a compact driving simulator developed by AVSimulation, is an out-of-the-box solution for non-racing research.

[0022] Savitsky-Golay is a commonly used polynomial smoothing and differential filtering algorithm. Its core function is to effectively filter out noise while preserving the trend characteristics of the data, and it can also directly calculate the higher-order derivatives of the data.

[0023] Step S2: Construct a three-dimensional risk factor model of "environmental complexity - task urgency - interference intensity"; Based on the data from step S1, a three-dimensional risk factor model of "environmental complexity - task urgency - interference intensity" is constructed. A four-level risk level key test scenario library is then built based on these risk factors. The risk factors for environmental complexity include weather type and traffic flow; the risk factors for task urgency include driving task type, minimum headway (THW), and minimum time to collision (TTC); and the risk factors for interference intensity include stimulus mode and interference duration. The four risk levels are extremely low risk, low risk, medium risk, and high risk, respectively. The classification criteria and level definitions for each dimension of the risk factors are as follows: The environmental complexity risk factor module categorizes risk factors into weather type and traffic flow based on two dimensions: natural environment and traffic environment. The weather type and risk level classifications are as follows: clear skies with visibility ≥ 20 km are defined as extremely low risk; cloudy skies with visibility 10 km-20 km are defined as low risk; rainy skies with visibility 5 km-10 km are defined as medium risk; and foggy or snowy skies with visibility ≤ 5 km are defined as high risk. The traffic flow risk level classifications are as follows: traffic flow ≤ 10 vehicles / km is defined as extremely low risk; traffic flow 10-20 vehicles / km is defined as low risk; traffic flow 20-30 vehicles / km is defined as medium risk; and traffic flow > 30 vehicles / km is defined as high risk.

[0024] The risk factors in the task urgency factor module are based on a two-dimensional factor of driving type and driving task safety regulations, specifically divided into three risk factors: driving task type, minimum headway (THW), and minimum time to collision (TTC). The risk level classification of the driving task type specifically includes: straight-line constant speed driving scenario is defined as extremely low risk level; smooth turning or following scenario is defined as low risk level; lane changing and obstacle avoidance behavior scenario is defined as medium risk level; emergency braking and collision warning scenario is defined as high risk level.

[0025] The minimum headway (THW) risk level classification specifically includes: scenarios with THW ≥ 5s are defined as extremely low risk; scenarios with THW 3s-5s are defined as low risk; scenarios with THW 2s-3s are defined as medium risk; and scenarios with THW ≤ 2s are defined as high risk. The minimum collision time (TTC) risk level classification specifically includes: scenarios with TTC ≥ 8s are defined as extremely low risk; scenarios with TTC 5s-8s are defined as low risk; scenarios with TTC 3s-5s are defined as medium risk; and scenarios with TTC ≤ 3s are defined as high risk.

[0026] The risk factors in the interference intensity risk factor module are specifically divided into stimulus modality and interference duration. The risk level classification of the stimulus modality specifically includes: no stimulus scenario is defined as extremely low risk level; single visual stimulus scenario is defined as low risk level; visual-auditory composite stimulus scenario is defined as medium risk level; visual-auditory-tactile multimodal composite stimulus scenario is defined as high risk level.

[0027] The risk level classification for interference duration includes: no interference scenarios are defined as extremely low risk; interference duration of 1s-3s is defined as low risk; interference duration of 3s-5s is defined as medium risk; and interference duration of ≥5s is defined as high risk.

[0028] Step S3: Setting up key test scenarios; The key test scenarios were set using orthogonal experimental design to determine the combination schemes of each risk factor, ensuring coverage of all four levels of risk scenarios. Specifically, this involved: first, orthogonally combining the sub-items of similar risk factors to achieve both randomness and scientific rigor; then, selecting scenarios from the orthogonally combined risk factors and screening four basic scenarios based on risk weights; finally, orthogonally combining the basic scenarios of the three risk factor categories again to form 4×4×4=64 basic scenarios. Based on the analytic hierarchy process (AHP), the key weight influencing factors of medium-risk and high-risk scenarios are more significant for system performance evaluation. Therefore, the top 20 scenarios, arranged in reverse order of risk level from the 64 basic scenarios, were selected as the key core scenarios of this invention.

[0029] Step S4: Virtual simulation modeling; Virtual simulation modeling: Based on the SCANeRStudio2025 and Prescan co-simulation platform, a virtual simulation scene and a high-fidelity dynamic model were constructed. The virtual simulation scene simulates an urban road with six lanes in both directions, each lane 3.5 meters wide, with double yellow lines in the middle and green belts on both sides, all of which are non-crossable areas. At the same time, static environmental elements were added to ensure the realism of the scene, including lawns, trees and houses on both sides of the open space. Scenes involving risk factors, such as weather type, traffic flow and driving task, were added selectively based on orthogonal experiments.

[0030] Among them, SCANeR Studio 2025 is a new generation simulation test platform for autonomous driving and advanced driver assistance systems (ADAS) launched by the French company AVSimulation.

[0031] Prescan is a professional-grade simulation testing platform for autonomous driving (AD) and advanced driver assistance systems (ADAS).

[0032] The high-fidelity dynamics model specifically includes the Savitsky-Golay filtering method that integrates the aforementioned scenario construction and the risk scenario level coupling method of the aforementioned three-dimensional risk factor model. The high-fidelity model relies on the simulation platform to accurately reproduce the vehicle's dynamic characteristics and to simulate extreme working conditions in a scenario-based manner, ultimately serving the performance evaluation of the brain-controlled vehicle system in a real driving environment.

[0033] Step S5: Real-world scenario reproduction; Real-world scenario reproduction: By modifying the actual vehicle site, a controllable water / snowmaking system is installed, and dynamic obstacles are set up to achieve a one-to-one mapping between the "simulation scenario and the actual vehicle scenario".

[0034] Step S6: Multi-dimensional performance evaluation indicator system; A multi-dimensional performance evaluation index system with a three-layer structure—"Target Layer - Overall Index Layer - Index Layer"—is constructed. The target layer provides multi-dimensional performance evaluation of the brain-controlled vehicle system; the overall index layer covers six dimensions: driving safety, anti-interference capability, generalization ability, timeliness, comfort, and data protection; the index layer comprises specific quantitative indicators under each overall index. Taking driving safety as an example, the index layer specifically includes TTC, THW, LCO, and EBD. The indicators for each dimension and their calculation methods are as follows: Driving safety indicators are used to evaluate the safety performance of brain-controlled vehicle systems during driving. This invention selects minimum time to collision (TTC), minimum headway (THW), lane centerline offset (LCO), and emergency braking distance (EBD) as evaluation indicators for driving safety.

[0035] Time to Collision (TTC) is a commonly used indicator for evaluating driving safety. It is often used to assess the safety risks of a vehicle while it is in motion. The smaller the value, the less safe the vehicle is. When TTC < 0.5s, a collision will occur. When TTC < 2.9s, a collision warning should be triggered. Therefore, the TTC safety threshold is set to ≥ 2.9s.

[0036] ; In the formula, For relative distance, For velocity (absolute value when relative velocity is negative); The absolute speed of a vehicle in a continuous series of vehicles; The speed of the vehicle targeting the obstacle.

[0037] Time Headway (TH): A crucial indicator of vehicle driving safety, TH is closely related to traffic flow composition and driving behavior, reflecting to some extent the capacity and service level of a road. Time Headway refers to the time difference between the front ends of consecutive vehicles passing the same point. It is calculated as the ratio of the distance between the front ends of consecutive vehicles to the speed of the following vehicle. This value represents the time it takes for the driver to adjust vehicle operations or react when the vehicle in front suddenly stops due to a malfunction or accident. ; In the formula, This refers to the headway between vehicles traveling in a continuous line.

[0038] In actual driving, the critical risk threshold of THW is 1.5s. When THW is 1.5 seconds, it means that the driver of the following vehicle has almost no reaction time to deal with the emergency braking of the vehicle in front. When THW < 2.5s, a collision warning needs to be triggered. Therefore, the THW safety threshold is set to ≥ 2.5s.

[0039] Emergency Braking Distance (EBD): It refers to the total distance traveled by a vehicle from when the driver starts to realize the need for emergency braking until the vehicle comes to a complete stop when encountering sudden danger. It is one of the core indicators for measuring the braking safety of a vehicle. Under the urban road commuting condition with a vehicle speed ≤ 50 km / h, the EBD safety threshold ≤ 30 m; under the condition of an ice and snow road surface with an adhesion coefficient of 1 / 3 - 1 / 2 of that of a dry asphalt road surface, the EBD safety threshold ≤ 50 m; under the highway condition with a vehicle speed ≤ 120 km / h, the EBD safety threshold ≤ 120 m.

[0040] ; In the formula, is the system response time, is the braking deceleration, is the initial braking speed.

[0041] Lateral Center Offset (LCO): It refers to the lateral distance between the vehicle center line and the lane center line. It is an important indicator for measuring the lateral position of a vehicle when driving within a lane and is often used to evaluate the driving stability and driving safety of a vehicle. On structured roads such as highways, LCO ≤ 0.3 m is a low risk; 0.3 m < LCO ≤ 0.8 m is a medium risk; LCO > 0.8 m is a high risk.

[0042] ; In the formula, is the lateral coordinate of the vehicle's center of mass (perpendicular to the driving direction) collected by an in-vehicle IMU or lidar, with the unit ; is the lateral coordinate of the lane center line obtained through SCANeR scene modeling or in-vehicle high-precision maps, with the unit ; An ideal brain-vehicle interaction system should have a certain degree of anti-interference ability, that is, when certain uncertain situations occur in the outside world, it can maintain the completeness of the acquired information and a certain execution efficiency, and can accurately identify the driver's intention. The anti-interference index is used to evaluate the stability of the brain-controlled vehicle system in a complex interference environment. In this invention, the recognition accuracy rate of the expected function of brain-vehicle interaction (ERA), false trigger rate (FPR), and signal-to-noise ratio (SNR) are selected as the evaluation indexes of anti-interference.

[0043] An ideal brain-vehicle interaction system should possess a certain degree of anti-interference capability, meaning that when certain uncertainties occur in the external environment, it should maintain the completeness of information acquisition and a certain level of execution efficiency, and be able to accurately identify the driver's intentions. Anti-interference metrics are used to evaluate the stability of the brain-controlled vehicle system in complex interference environments. This invention selects the expected function recognition accuracy (ERA), false trigger rate (FPR), and signal-to-noise ratio (SNR) as evaluation metrics for anti-interference capability.

[0044] Error Recognition Accuracy (ERA): This is the percentage of times a driver's brainwave intent is correctly recognized. It is an indicator used to evaluate the accuracy of a vehicle's function recognition system. It represents the proportion of times the system correctly recognizes function-related information. For safety-critical functions such as collision warning and emergency braking, the ERA safety threshold is ≥95%; for non-safety-critical functions such as window operation and air conditioning adjustment, the ERA safety threshold is ≥90%.

[0045] ; In the formula, To correctly identify the number of times, This represents the number of incorrect identifications.

[0046] False Positive Rate (FPR): The probability of issuing an incorrect control command in an idle state. It is a key indicator for measuring the anti-interference capability of a vehicle safety system, referring to the probability that the system will incorrectly issue a trigger signal in a scenario without real risk. In normal driving scenarios on highways under clear weather, the FPR safety threshold is ≤0.1%; in urban road scenarios, the FPR safety threshold is relaxed to ≤0.5%.

[0047] ; In the formula, Number of false positives The number of true negatives.

[0048] Signal-to-Noise Ratio (SNR): In the field of vehicle driving safety, SNR is the ratio of the amplitude of the EEG signal to the amplitude of the interference signal. It is a key indicator for measuring the ratio of the intensity of "effective driving signals" to "environmental / system interference signals" and is a core dimension of interference resistance evaluation. Under daytime heavy rain conditions on highways, the SNR safety threshold is ≥15dB; under ordinary urban road conditions, the SNR safety threshold is ≥10dB.

[0049] ; In the formula, This represents the effective value of the signal amplitude. The effective value representing the amplitude of the signal noise. alpha wave (8-13Hz) amplitude, This represents the amplitude of electromyographic interference.

[0050] Generalization metrics are used to evaluate the adaptability of brain-controlled vehicle systems to different subjects and scenarios. In scenarios such as driving behavior analysis, autonomous driving system testing, or driver state monitoring, "generalization" is a core indicator for measuring the practicality of a model or system, that is, the system's performance on "new objects outside of the training data". This invention selects cross-subject accuracy (CPA) and cross-scenario suitability (CSA) for brain-controlled vehicle interaction expected function recognition as evaluation metrics for generalization.

[0051] Cross-Participant Accuracy (CPA): CPA is the average recognition accuracy when different subjects use the system. It refers to the average functional recognition accuracy of the system on a new group of subjects who have not participated in the training, and is used to quantify the system's suitability for different drivers. The basic pass threshold is CPA ≥ 80%; the excellent threshold is CPA ≥ 85%, which meets the multi-user scenario requirements of L2+ level driver assistance systems.

[0052] ; In the formula, For the number of subjects, For the first The accuracy rate of the subjects.

[0053] Cross-Scenario Adaptation Rate (CSA): CSA is the average functional achievement rate of a system under different risk scenarios. It refers to the proportion of functions that perform adequately in the "source scenario" (the benchmark scenario where the system was trained / calibrated) and can still meet the functional performance requirements after being migrated to the "target scenario" (such as a new scenario where the source scenario differs in environment and road conditions). In low-difference scenarios (e.g., sunny urban area → cloudy urban area), CSA ≥ 90%; in medium-difference scenarios (e.g., sunny urban area → rainy urban area, or urban area with streetlights at night), CSA ≥ 80% (rainy day) and CSA ≥ 85% (night); in high-difference scenarios (e.g., sunny urban area → heavy rain with no streetlights), CSA ≥ 70%.

[0054] ; In the formula, For the number of scenes, For the scene Number of times the function was successfully implemented in China This represents the total number of tests.

[0055] Timeliness indicators are used to evaluate the response speed and information transmission efficiency of the brain-controlled vehicle system. This invention selects the response time (RT), information transmission rate (ITR), and instruction execution delay (ED) of the brain-controlled vehicle interaction expected function recognition instruction as evaluation indicators of timeliness.

[0056] Response Time (RT): In Advanced Driver Assistance Systems (ADAS) or autonomous driving systems, RT is the time difference between the driver issuing an intention and the execution of the function. It is a core indicator for measuring the system's timeliness, referring to the time interval from "the system receiving the trigger command" to "the system executing the corresponding action." Safety thresholds: In high-speed cruising scenarios, RT ≤ 0.5s for normal situations and RT ≤ 0.3s for emergency scenarios; in urban road scenarios, RT ≤ 0.8s.

[0057] ; In the formula, To record the timing of EEG intention triggering based on the TSN time synchronization system, This refers to the timing of command execution by the vehicle ECU based on the TSN time synchronization system.

[0058] Information Transfer Rate (ITR): In Advanced Driver Assistance Systems (ADAS) or intelligent driving scenarios, IRT is the number of effective bits of information transmitted by the system per unit time. It is a core timeliness indicator for measuring the amount of critical driving information effectively transmitted between the system and the driver / other vehicles / roadside equipment per unit time. Safety threshold: In highway scenarios, basic IRT ≥ 8 bits / s, total critical information transmission time ≤ 0.3s, and redundant information ratio ≤ 10%.

[0059] ; In the formula, For single selection of time, For the number of optional targets, For accuracy.

[0060] Execution Delay (ED): ED refers to the time difference between the system outputting an instruction and the vehicle's hardware response. It is one of the core indicators for measuring timeliness, referring to the time interval from when the system receives the "instruction signal to be executed" to when the instruction is fully executed. Safety thresholds: Under normal high-speed operation, ED ≤ 0.5s; under high-speed emergency avoidance, ED ≤ 0.3s; under low-speed urban scenarios, ED ≤ 1s.

[0061] ; In the formula, Record the CANFD bus command transmission time. The moment when a vehicle actuator (such as a window motor) operates.

[0062] Comfort indicators are used to evaluate the driver's experience using the brain-controlled vehicle system. This invention selects the mental workload (ML) of brain-vehicle interaction expectation function recognition, device wearing comfort (WC), and operation satisfaction (OS) as comfort evaluation indicators.

[0063] Mental Load (ML) is a core indicator for measuring the cognitive stress a driver experiences when receiving and processing driving-related information. It directly impacts driving comfort, concentration, and the efficiency of safe decision-making. The ML threshold needs to be set in conjunction with the driving scenario. The core principle is to ensure that the driver maintains ML without experiencing significant fatigue. Currently, based on numerous driving experiments, the industry has established the following general threshold standards, using a 0-10 scoring system as an example: In typical driving scenarios, ML ≤ 3 points indicates a very low load, corresponding to a driving state of inattention and slow reaction time, such as ML=2 on an open highway when the driver is using a mobile phone; ML 3-7 points indicates a comfortable load, corresponding to a driving state of focused concentration and cognitive ease; ML > 7 points indicates a very high load, corresponding to a driving state of cognitive fatigue and decision-making errors. The NASA-TLX scale is used to calculate a weighted average score across six dimensions: mental demand, physical demand, time demand, effort level, frustration, and performance. ; In the formula, For dimension weights, The dimension score is 0-100, where n is the core component dimension of ML.

[0064] Wearing Comfort (WC): WC is a core indicator for measuring driver comfort when wearing assistive devices such as smart helmets, eye-tracking glasses, heart rate monitoring wristbands, and in-vehicle voice interaction headsets. Wearing Comfort (WC) is not a single indicator but a comprehensive evaluation that integrates physiological discomfort such as localized pressure, impaired blood circulation, sweating and stuffiness, psychological feelings, and driving suitability. Taking congested urban roads as an example, the overall WC score should be ≥0.7. If WC <0.7, the driver should adjust the device ≥3 times every 10 minutes. There should be no "continuous pressure" or "stuffiness." If continuous pressure is present, the driver will frequently look down to adjust the device. Objectively, after one hour of wear, there should be no pressure marks on the skin and sweating should be ≤6g / m². Skin pressure marks or excessive sweating can lead to stiff hand movements and delayed braking.

[0065] ; In the formula: The overall score for the comfort of wearing the device is given, ranging from 0 to 1. The closer the score is to 1, the higher the comfort level. Driver comfort was rated using the internationally recognized "9-point comfort scale". That is, "1 = extreme discomfort, 9 = extreme comfort"; This is an objective physiological indicator, expressed as local pressure force, with the unit being kPa, such as the pressure exerted by a helmet on the forehead. Pa levels exceeding 5 kPa can easily obstruct blood circulation. This is an objective physiological indicator, expressed as changes in skin moisture, with units of g / m², such as the amount of sweat produced in the area where the device is worn on the wrist. ,Exceed It will produce a noticeably sticky feeling; As an objective physiological indicator, it is expressed as muscle tension, with the unit being μV. It is measured using an electromyography (EMG) sensor in the neck / hand muscles. ,Exceed This indicates that the muscles are tense due to discomfort with the equipment.

[0066] In driving scenarios, Operation Satisfaction (OS) is a core indicator measuring the driver's satisfaction with the operation process and results of the vehicle control system. It directly affects driving comfort, driving confidence, and long-term user experience. It is not only related to the rationality of the device's physical design but also closely related to functional characteristics such as operational response speed, accuracy, and fault tolerance. It is one of the key dimensions for evaluating the quality of vehicle human-machine interface (HMI) design. In general scenarios, the core safety threshold for OS, on a 0-100 scale, is an overall OS ≥ 70 points, with OS for key operations such as braking, steering, and accelerator ≥ 80 points. If OS < 70 points, it is considered a "poor operating experience." A key operation OS < 80 points can lead to the driver "not braking in time" or "not applying the correct pressure," increasing the risk of collision. In urban congested traffic scenarios, the scenario adaptation threshold is a key operation OS ≥ 85 points and an auxiliary operation OS ≥ 75 points. If the accelerator operation OS < 85 points, the driver will frequently adjust their footwork due to "unpredictable accelerator response," leading to ankle fatigue and distraction in maintaining following distance.

[0067] ; In the formula, OS is the total operation satisfaction score, which usually ranges from 0 to 100 points. The higher the score, the higher the satisfaction. The weight of the i-th evaluation dimension is set according to the importance of the scenario, and the sum of the weights of all dimensions is 1. For example, the weight of "operational accuracy" is 0.4, the weight of "process comfort" is 0.3, and the weight of "fault tolerance" is 0.3. The score for the i-th evaluation dimension ranges from 0 to 10. It is calculated using driver questionnaires and data. For example, the score for "operational accuracy" is calculated as: (number of times the actual operation matches the expectation / total number of operations) × 10. n represents the number of evaluation dimensions.

[0068] Data security is used to evaluate the ability of the brain-controlled vehicle system to protect the driver's sensitive brainwave data. This invention selects the data encryption strength (ES), data transmission packet loss rate (PLR), privacy data anonymization rate (PDR), and data storage integrity (ESI) as evaluation indicators for data security.

[0069] Encryption Strength (ES): In intelligent connected vehicles (ICVs) or advanced driver assistance systems (ADAS), ES is a core security indicator that measures the protection of vehicle data against unauthorized access, tampering, or theft during transmission and storage. It quantifies the complexity of encryption algorithms, key length, and resistance to cracking to ensure that driving data is not maliciously attacked, thereby avoiding driving safety risks caused by data leakage or tampering. In urban road vehicle-to-vehicle communication (V2V) conditions, the security threshold for ES (Electronic Security Detection) is ≥7 points. The specific classifications and corresponding risks are as follows: An ES score of 8-10 points indicates a high security level, with extremely strong resistance to hacking and the ability to withstand national-level computing power attacks. It ensures data authenticity and integrity, and there are no security risks in system decision-making. An ES score of 7-8 points represents the security threshold, capable of resisting conventional hacker attacks, ensuring data transmission and storage security; only extreme computing power attacks could lead to risks. An ES score of 5-7 points indicates a low security level, easily cracked by professional hackers, potentially leading to data tampering and increased risk of collision warning delays. An ES score <5 points indicates an insecure level, with no effective encryption or easily cracked encryption, making data highly susceptible to tampering. This could cause the system to misjudge vehicle distances, miss collision warnings, and lead to accidents.

[0070] ; In the formula, ES is the final data encryption strength score, with a value range of [0,100]. The higher the score, the stronger the encryption protection capability. A score of 80 or above is the security compliance line for ICV / ADAS scenarios. , , Weight coefficients for each dimension, satisfying The weights need to be dynamically adjusted based on the specific sub-scenario of ICV / ADAS. The encryption algorithm is scored in terms of its resistance to cracking complexity [0, 100], which is the core measure of the algorithm's ability to resist brute-force attacks and differential attacks; The security score for the entire lifecycle management of keys is [0, 100], which measures the protection capabilities of key generation, storage, update, and destruction. The adaptability score for data transmission / storage scenarios is [0, 100].

[0071] In vehicle-to-everything (V2X) or intelligent driving systems, packet loss rate (PLR) refers to the percentage of data packets lost during transmission and is one of the core indicators for measuring data security. In the transmission of EEG data between the EEG acquisition device, the vehicle ECU, and the cloud server, the proportion of lost data packets out of the total transmitted data packets is used to quantify data transmission reliability. The PLR ​​safety threshold needs to be set in conjunction with the real-time requirements of the driving scenario: the thresholds differ significantly for highway scenarios, urban congestion scenarios, and suburban open scenarios. This study selects urban roads with high vehicle density, frequent lane changes, and numerous unexpected scenarios for V2X operations, where THW and TTC have the highest adjustment frequency as typical scenarios. Safety thresholds are given as follows: In urban roads with high vehicle density, frequent lane changes, and numerous unexpected scenarios, a PLR (Power Loss Rate) of 0% to 0.5% represents an extremely safe level, with almost no data loss, calculation error <1%, and warning response delay <10ms, fully meeting safety decision-making requirements; a PLR of 0.5% to 2% represents a safe level, where occasional packet loss can be compensated for by a "data retransmission mechanism," with calculation error <3% and warning delay <50ms, posing no safety risk; a PLR of 2% to 5% represents a critical risk level, where packet loss frequency increases and the retransmission mechanism cannot fully cover it, with calculation error reaching 5% to 8% and warning delay exceeding 100ms, requiring a "speed reduction reminder" to be triggered; a PLR > 5% represents a high-risk level, where significant data loss leads to calculation distortion and warning failure, directly increasing the risk of collision, requiring the triggering of an "emergency avoidance mode," such as active deceleration and activating hazard lights.

[0072] ; In the formula, For the number of lost packets, This represents the total number of data packets.

[0073] Among them, the urban road V2X (vehicle-to-infrastructure) working condition includes four types of interaction scenarios: vehicle-to-vehicle (V2V), vehicle-to-infrastructure (V2I), vehicle-to-person (V2P), and vehicle-to-cloud (V2N). It primarily addresses issues such as perception blind spots, signal coordination, and risk warning in complex urban traffic environments, and is a key support for autonomous driving systems to improve the safety and efficiency of urban road traffic.

[0074] Privacy Data Desensitization Rate (PDR): In scenarios such as vehicle-to-everything (V2X) and autonomous driving, vehicles continuously collect and transmit large amounts of data containing privacy information. PDR refers to the proportion of sensitive information in EEG data that has been desensitized. It is a core indicator that measures the degree to which this privacy data is effectively desensitized during processing, storage, and transmission, directly determining the risk of privacy data leakage and serving as a key dimension for data security. The security threshold for PDR needs to be determined in conjunction with the "degree of harm of privacy data leakage" in specific driving scenarios. Among these, autonomous driving on urban roads is a scenario with high privacy risks, requiring a PDR ≥ 95%. Specifically, the desensitization rate of key features involved in identity recognition must reach 100% to prevent reverse identification of the driver's identity through EEG data.

[0075] ; In the formula, This refers to the amount of sensitive data that has been anonymized. This represents the total amount of sensitive data.

[0076] Data Storage Integrity (DSI): DSI refers to the consistency between stored data and the original data. It is one of the core indicators of data security. Its core function is to measure whether "critical data generated or received during vehicle operation remains in its original state, without tampering, loss, or damage" within the storage period. A DSI ≥ 99.9% indicates a safe level, with extremely high core data integrity, accurate calculations, and no security risks. A DSI between 99.0% and 99.9% indicates a warning level, with a small amount of non-core data missing or damaged, causing slight deviations in calculation accuracy. A DSI < 99.0% indicates a dangerous level, with core data lost or tampered with, leading to severe calculation distortions and potentially causing accidents such as collisions or rear-end collisions, requiring immediate downgrade to manual driving.

[0077] ; In the formula, Total amount of data stored To determine the total amount of data to be stored, the DSI value range is: .when When the value is 1, it means that all stored data has passed the integrity check and there is no damage; when the value is 1, it means that all stored data has passed the integrity check and there is no damage. When this occurs, it means that all stored data is corrupted or lost, resulting in a complete loss of integrity.

[0078] Step S7: Construction of a closed-loop virtual-real testing platform integrating "virtual simulation, real vehicle testing, data synchronization, and data security"; The specific architecture is as follows: The virtual simulation subsystem generates four-level risk test scenarios, outputs scenario parameters and virtual vehicle status, and synchronizes scenario commands to the real vehicle subsystem. The subsystem consists of hardware and software systems. The hardware system specifically includes industrial-grade servers, such as an Intel Xeon Gold 6426R CPU, an NVIDIA A100 GPU, 256GB DDR5 memory, and a 4K triple-screen projection system. The software system specifically includes the scenario modeling software SCANeR Studio2025, the vehicle dynamics simulation software Prescan 2024, and the EEG signal simulation software MATLAB / Simulink. The real-vehicle testing subsystem is based on a modified Hongqi H9 experimental vehicle. The original driving control module was removed, and a brain-controlled interface module was added, compatible with the CANFD bus. The subsystem specifically includes an EEG acquisition module, a vehicle status acquisition module, and a stimulus generation module. The EEG acquisition module uses a Neuroscan SynAmps RT multi-channel EEG analyzer: 64 channels, 1000Hz sampling rate, input impedance ≥10GΩ, and Ag / AgCl electrodes to collect information from key brain regions, specifically covering Fpz, F3, F4, P3, P4, O1, and O2. The vehicle status acquisition module is equipped with millimeter-wave radar (detection range 0.1-200m, accuracy ±0.1m); lidar (128 lines, point cloud density 200 points / m²); and an IMU (1000Hz sampling rate, acceleration accuracy ±0.01m / s²). The stimulation generation module is mainly equipped with an in-vehicle display screen for visual cues, an in-vehicle audio system for auditory cues (frequency 500-1000Hz), and a seat vibration module for tactile cues (vibration frequency 20-50Hz, amplitude 0.5-2mm).

[0079] The data synchronization subsystem can achieve time alignment of multi-source data from virtual simulation, real vehicles, and EEG, specifically including synchronization protocols, synchronization nodes, and unified data format processing.

[0080] The synchronization protocol specifically adopts 5G+Time Sensitive Network (TSN) and supports nanosecond-level time synchronization. The synchronization node is achieved by deploying three time synchronization nodes: a scenario node for the simulation subsystem, an EEG node for the EEG acquisition module, and a vehicle node for the real vehicle subsystem. All nodes are connected via a TSN switch. The data is uniformly encapsulated in JSON-LD format, which includes timestamp UTC time and local nanosecond offset, data type, specifically including scene / EEG / vehicle, and data content such as EEG amplitude and vehicle speed.

[0081] Data security specifically protects sensitive biometric information, primarily EEG data. Through a comprehensive design of "transmission encryption - real-time verification - desensitization processing - storage backup," it provides complete integrated security protection for the driver's EEG privacy data, deeply coupled with virtual simulation, real vehicle testing, and data synchronization subsystems.

[0082] Step S8: Establish standardized evaluation methods; Experimental Preparation Phase: Participants must meet the following screening criteria: age 25-40 years, driving experience ≥3 years with a C1 or higher driver's license, no brain or mental illness, normal vision or corrected vision ≥1.0, and pass the Mini-Mental State Examination (MMSE) with a score ≥27, indicating normal cognitive ability. Training includes a 2-hour pre-experiment session, specifically covering EEG device wearing training to ensure electrode impedance ≤5kΩ for accurate EEG signal acquisition, brain-controlled function training (e.g., using "imagining a left fist to control the brake" and "imagining a right fist to control the window"), and scenario adaptation training to familiarize participants with the operational logic of different risk levels. After training, participants must achieve a brain-controlled function recognition accuracy rate ≥80% to proceed to the formal experiment.

[0083] Experimental Design and Variable Control: Based on the risk level of each scenario, the experiment was divided into four groups: very low risk, low risk, medium risk, and high risk. Each group contained five core scenarios, testing five brain-controlled functions: window operation, air conditioning temperature adjustment, music switching, volume adjustment, and braking. Variable control required clearly defining the test scenarios at different risk levels to cover diverse driving situations, setting geographical attribute parameters based on the actual driving environment, standardizing the participants' brain-computer interaction behavior to ensure experimental consistency, and simulating the influence of different natural conditions on the test environment. For speed control, the speeds were 15-30 km / h for the very low risk group, 10-25 km / h for the low risk group, 8-20 km / h for the medium risk group, and 5-15 km / h for the high risk group. Stimulus types included weak stimuli such as dashboard light prompts with a brightness of 500 cd / m², and strong stimuli such as lights + 80 dB alarm sound + seat vibration with an amplitude of 1 mm. Each scenario and function combination was tested 10 times to ensure statistical significance of the experimental data, with a sample size ≥ 50.

[0084] The variable control is achieved through four dimensions: scenario risk grading, geographical parameter standardization, behavioral norm constraints, and natural condition simulation, ensuring the repeatability of experimental data and the reliability of results. The scenario risk grading is based on a three-dimensional risk factor model of "environmental complexity - task urgency - interference intensity," dividing test scenarios into four risk levels: extremely low, low, medium, and high, covering all operating conditions from normal driving to extremely dangerous situations. The geographical parameter standardization standardizes geographical parameters through virtual simulation and real-vehicle testing, specifically including the actual simulation settings of road structure, traffic flow, and obstacles to simulate a real driving environment. The behavioral norm constraints specifically include standardized subject selection and standardized training processes. The natural condition simulation simulates weather, lighting, and multimodal interference through multiple means to test the system's stability in complex environments, specifically including simulations of different weather conditions and visual, auditory, and other multimodal interferences.

[0085] The experimental steps for the formal experiment are as follows: Step S81: Upon arrival at the laboratory, the subject first fills out an informed consent form. Then, the EEG device is fitted, and electrode impedance calibration is performed to ensure that all channel impedances are ≤5kΩ, guaranteeing normal EEG signal acquisition. Step S82: Start the test platform and synchronize the virtual simulation subsystem, the real vehicle subsystem, and the data synchronization subsystem to confirm that the communication of each module is normal. For example, the EEG data sampling rate is stable at 1000Hz and there is no packet loss on the vehicle CANFD bus.

[0086] Step S83: Conduct tests sequentially according to the scenario risk level from low to high to avoid affecting the test results of high-risk scenarios due to subject fatigue. Before each scenario begins, the simulation subsystem will output scenario prompts, such as "Medium-risk scenario: Rainy day + obstacle avoidance task + strong stimulus", so that the subject understands the scenario information.

[0087] Step S84: During the scenario execution, the subject executes corresponding brain-controlled functions based on stimulus cues. For example, upon receiving a "brake" stimulus, the subject triggers the brakes by imagining making a fist with their left hand. During this process, the data synchronization subsystem collects EEG data, vehicle data, and scenario data in real time.

[0088] After each risk group test, participants were given a 15-minute rest period to prevent mental fatigue from affecting subsequent experiments. During the rest period, participants completed a 5-question simple comfort questionnaire to assess their immediate wearing experience. The total duration of all scenario tests was approximately 3 hours. Participants were required to complete three tasks: first, complete the NASA-TLX Mental Load Scale; second, complete a 20-question functional satisfaction questionnaire using a Likert scale; and third, participate in a semi-structured interview to record their suggestions for system improvement.

[0089] The NASA-TLX is a globally recognized mental workload assessment scale released by NASA's Ames Research Center in the 1980s.

[0090] Data processing comprises two parts: data preprocessing and data backup. In data preprocessing, the Savitsky-Golay filter with a window length of 5 and a polynomial order of 2 is used to remove jitter for vehicle data, while an 8Hz high-pass filter and a 30Hz low-pass filter are used for EEG data to remove 50Hz power frequency interference and EMG noise, ensuring data quality. In the data backup stage, experimental data is stored in real time on edge computing nodes equipped with encrypted hard drives and uploaded to the cloud distributed storage Alibaba Cloud OSS within 24 hours. At the same time, a SHA-256 hash value is generated for data integrity verification to prevent data loss or tampering.

[0091] This completes the establishment of standardized evaluation methods.

[0092] Example: Based on the established three-dimensional risk model and multi-dimensional evaluation system, and relying on the brain-computer interface (BCI) expected function testing device, this experiment, with window control as the core test scenario, recruits drivers to conduct BCI experiments, aiming to evaluate the performance of the brain-computer interface system. The experimental flowchart is attached. Figure 4 As shown, the specific testing plan and the working method of the brain-computer interface system are as follows: The test scenarios cover five road types: straight, turning, T-junction, crossroads, and roundabouts. The test vehicle travels at a constant speed of 10-30 km / h along the lane, with a traffic flow set at 20 vehicles / km, and the test is conducted under rainy conditions in the simulator. The brain-computer interface system works as follows: When the test vehicle reaches the designated location, the driver needs to generate a window control intention through motor imagery. This intention is collected by the brain-computer device and transmitted to the industrial control computer. The industrial control computer decodes the signal and converts it into a corresponding control command, which is then sent to the vehicle control system to execute the operation of lowering or raising the window. At the same time, when the vehicle has traveled 2 seconds into the designated location, a interference scenario is created by visual stimulation (flashing brake malfunction light on the dashboard) and auditory stimulation (emergency broadcast on the in-car radio). When the vehicle has traveled 5 seconds into the designated location, the driving task is changed, and lane changing or obstacle avoidance behaviors are adopted to simulate a complex driving environment. During the test, the system simultaneously records timestamps, driver EEG data, and brain-computer interface response time. The test ends after the vehicle completes the window raising and lowering test on five different road types and reaches the finish line.

[0093] Based on the definition criteria for medium-risk scenarios using the three-dimensional risk factor model of "environmental complexity - task urgency - interference intensity," and combined with the characteristics of real driving environments, this invention is specifically applied using "air conditioning lift function evaluation in medium-risk scenarios" as an example: Environmental complexity parameters: The weather type is set as a rainy environment according to the medium-risk weather definition standard in the patent, and the visibility is controlled within the range of 5km-10km; natural rainfall is simulated by a real vehicle on-site water spraying system, and the rainfall is set as a moderate rain intensity of 2.5-10mm / h to ensure that the road surface friction coefficient is maintained between 0.4-0.6, which meets the requirements of road surface dynamic characteristics under medium-risk scenarios; Traffic flow was categorized into risk levels of 20-30 vehicles / km. Typical sections of urban secondary arterial roads were selected as the scenario prototype. Traffic flow was generated using the virtual simulation subsystem SCANeR Studio 2025. Cars accounted for 97.02%, motorcycles for 0.7%, and trucks for 2.28%. Vehicle trajectories were based on real trajectory data from the Lankershim Boulevard and Peachtree sections in the NGSIM dataset. After denoising using a Savitsky-Golay filter with a window length of 15, the data was imported to ensure that the traffic flow movement conformed to real driving patterns.

[0094] Task urgency parameters: The driving task type is set as a lane changing task. The subject needs to complete one lateral lane change operation during the scenario. The lane change is triggered when the speed of the vehicle in front is 10 km / h lower than the speed of the subject. During the lane change, the simulation subsystem controls the relative position of the vehicle in front and the subject to ensure that the minimum headway between the subject and the vehicle in front at the start of the lane change is 2s-3s and the minimum collision time is 3s-5s, which meets the medium-risk task urgency standard. The road type scenario includes straight road sections and curved road sections. The straight road section is 200m long and the curve radius is 50m, which conforms to the design specifications of urban secondary arterial roads. The road markings adopt standard urban road markings, the lane width is 3.5m, and a double yellow line is set in the middle that cannot be crossed. Green belts with a width of 1.5m are set on both sides. The static environmental elements of trees and houses are arranged in the top view layout of the driving simulator scenario to enhance the realism of the scene.

[0095] Interference intensity parameters: The stimulation modality adopts visual-auditory composite stimulation. The visual stimulation is achieved through dashboard light prompts, with the light brightness set at 500 cd / ㎡ and the flashing frequency at 2Hz. Auditory stimulation is output through the vehicle's audio system, with the sound type being a voice announcement "Lane change required ahead, please prepare," at a volume of 80dB. The stimulation trigger time is synchronized with the lane change preparation time, and the duration is 3s-5s. Two dynamic interference sources are set in the scenario: a non-motorized vehicle crossing laterally at a speed of 15km / h and a vehicle temporarily parked on the roadside for 10s. The locations of the interference sources are randomly distributed during the 30s-60s time period of the scenario operation to simulate sudden interference situations in real driving.

[0096] The specific steps for software deployment are as follows: The data synchronization subsystem adopts the 5G+TSN time-sensitive network synchronization protocol and deploys three time synchronization nodes: a scene node, an EEG node, and a vehicle node. It achieves nanosecond-level time synchronization accuracy of ≤100ns through a TSN switch. The data format is uniformly encapsulated in JSON-LD format, which includes UTC timestamps accurate to nanoseconds, data types (scene / EEG / vehicle), and data content such as EEG amplitude, vehicle speed, and relative distance, to ensure time alignment of multi-source data.

[0097] Data storage subsystem: Experimental data is stored in real time on edge computing nodes equipped with 2TB encrypted hard drives, using the AES-256 encryption algorithm, and uploaded to cloud distributed storage within 24 hours. At the same time, SHA-256 hash values ​​are generated for data integrity verification to prevent data loss or tampering.

[0098] The specific steps for vehicle data collection are as follows: Vehicle motion parameters: Speed, acceleration, lateral and longitudinal acceleration, and yaw angle of the vehicle are collected by IMU at a sampling rate of 1000Hz; control parameters such as brake pedal opening, accelerator pedal opening, steering wheel angle and speed are read by the vehicle CANFD bus at a sampling frequency of 500Hz; speed data is uniformly converted to m / s and acceleration is converted to m / s², with a conversion factor of 0.3048 feet-meters.

[0099] Relative position and status parameters: The relative distance and relative speed between the vehicle and the vehicles in front and beside it are collected using millimeter-wave radar at a detection frequency of 10Hz and a data accuracy of ±0.1m; the contour information of surrounding vehicles and lane line positions are obtained through lidar at a sampling rate of 10Hz, which is used to calculate the lane centerline offset (LCO); all vehicle data are filtered by Savitsky-Golay with a window length of 5 and a polynomial order of 2 to remove jitter before storage to ensure data stability.

[0100] The specific steps for collecting EEG data are as follows: Data acquisition equipment and channels: [Adopted] The multi-channel EEG analyzer has 64 channels, a sampling rate of 1000Hz, and an input impedance of ≥10GΩ. It is equipped with Ag / AgCl electrodes, and the acquisition electrodes cover key brain regions such as Fpz, F3, F4, P3, P4, O1, and O2. The electrode impedance is calibrated to ≤5kΩ to ensure the accuracy of EEG signal acquisition. The acquired EEG signals include alpha waves of 8-13Hz and beta waves of 14-30Hz, which are used for subsequent interference resistance index signal-to-noise ratio (SNR) calculation.

[0101] Acquisition Synchronization and Preprocessing: The EEG data acquisition is time-aligned with the vehicle data and scenario data through the TSN time synchronization system, and the timestamp error is ≤ 100 ns. The original EEG data is filtered with an 8 Hz high-pass filter + 30 Hz low-pass filter to remove the power frequency interference of 50 Hz and EMG noise, and then the eye and ECG artifacts are removed through independent component analysis (ICA). The preprocessed data is stored in the EDF format for subsequent analysis.

[0102] Calculation of Evaluation Indexes during the Execution of the Air Conditioner Lifting Function In the medium-risk test scenario, the subject triggers the temperature adjustment function of the air conditioner of the brain-controlled vehicle system through the action of "imagining clenching the right fist". The temperature adjustment range is 18°C - 28°C, and the adjustment step is 1°C. During the execution of the system function, based on the collected vehicle data, EEG data, and evaluation data, the relevant indexes of the multi-dimensional performance evaluation index system in the patent are calculated. The specific calculation process is as follows:

[0103] Index Comparison Results Compare each evaluation index during the execution of the air conditioner lifting function in the medium-risk scenario with the safety threshold. The specific results are as follows: Comparison of Driving Safety Indexes The calculated value of the minimum time headway (THW) is 2.4 s, which is within the range of 2 s - 3 s required by the medium-risk scenario and greater than the safety threshold of 2 s, meeting the driving safety requirements. The calculated value of the minimum time to collision (TTC) is 9.5 s, far greater than the upper limit of the 3 s - 5 s range required by the medium-risk scenario, with no collision risk and meeting the safety standards. The calculated value of the lane centerline offset (LCO) is 0.3 m, within the reasonable range of 0.3 m < LCO ≤ 0.8 m in the medium-risk scenario, not exceeding the safety threshold of 0.8 m, and the vehicle driving trajectory is stable. The calculated value of the emergency braking distance (EBD) is 5.08 m, far less than the safety threshold of 30 m for urban roads, and the vehicle braking performance is reliable in case of sudden emergencies.

[0104] Comparison of Anti-Interference Indexes The calculated value of the function recognition accuracy rate (ERA) is 90%, reaching the safety threshold of ≥ 90% for non-safety-critical functions, and the recognition accuracy of the system for the air conditioner lifting intention is qualified. The calculated value of the false trigger rate (FPR) is 2%, exceeding the safety threshold of ≤ 0.5%. It is necessary to reduce the false trigger probability by optimizing the EEG signal feature extraction algorithm (such as adding a wavelet packet decomposition denoising link). The calculated value of the signal-to-noise ratio (SNR) is 13.98 dB, greater than the safety threshold of 10 dB, indicating that the EEG signal is less affected by interference and the signal quality is good.

[0105] Comparison of timeliness indicators The command response time (RT) is calculated to be 0.45s, which is less than the urban road safety threshold of 0.8s. The system responds quickly to air conditioning lifting commands with no significant delay. The Information Transfer Rate (ITR) is calculated to be 0.106 bits / s, which is slightly higher than the non-security-critical function threshold of ≥0.1 bits / s, and the data transmission efficiency meets the basic requirements.

[0106] Comparison of comfort indicators The mental workload (ML) value was 52, which is within the reasonable range of 40-60. The subject did not experience excessive mental fatigue when performing the air conditioning lifting function. The device wearing comfort (WC) score is 3.8, which is greater than the qualified threshold of 3, indicating that the wearing experience of the EEG device meets the requirements. The Operational Satisfaction (OS) score was 5.5, which is above the passing threshold of 5, indicating that the participants had a high level of satisfaction with the operation process and effectiveness of the air conditioning lifting function.

[0107] Based on the comparison of various indicators with the safety threshold, the following verification conclusions are drawn: Overall performance is satisfactory: In a medium-risk test scenario, when the brain-controlled vehicle system performs the air conditioning lifting function, the driving safety, timeliness, and comfort indicators all meet the safety threshold requirements. The system can achieve stable adjustment of the air conditioning temperature while ensuring driving safety, and the test subjects have a good operating experience. Local optimization is needed: The false trigger rate (FPR) in the anti-interference index exceeds the safety threshold, which poses a risk that the system will falsely trigger the air conditioning lifting function when there is no control intention. The false trigger rate needs to be reduced by optimizing the algorithm (such as introducing an attention mechanism to enhance the distinguishability of EEG signal features) and adding an intention verification step (such as dual EEG feature matching) to further improve the system reliability. Good adaptability to different scenarios: Under medium-risk scenarios such as rain, moderate traffic flow, and complex interference, the system's various indicators can still be maintained within a reasonable range, indicating that the system has good adaptability to medium-risk driving environments and can meet the usage needs of typical medium-risk scenarios such as urban secondary arterial roads.

[0108] This verification demonstrates that the brain-controlled vehicle system performance evaluation and assessment method proposed in this invention can effectively evaluate the performance of the air conditioning lifting function in medium-risk scenarios, providing accurate data support and directional guidance for the optimization and upgrading of the brain-controlled vehicle system.

Claims

1. A performance evaluation and assessment method for brain-controlled vehicle systems oriented towards key testing scenarios, characterized in that, Includes the following steps: Step S1: Constructing the test scenario; By integrating a driving simulator and an EEG device, a cloud platform for synchronizing human-vehicle-environment information is constructed to simulate diverse road environments, driving tasks, and in-vehicle environmental interference scenarios. The platform can collect EEG signals, vehicle motion parameters, and driving scenario data in real time through the simulator and EEG device, achieving synchronous acquisition of multi-source data. At the same time, the Savitsky-Golay filtering algorithm is used to denoise the vehicle motion parameters. Step S2: Construct a three-dimensional risk factor model of environmental complexity, task urgency, and interference intensity; Based on the data from step S1, a three-dimensional risk factor model of environmental complexity, task urgency, and interference intensity is constructed. A key test scenario library with four risk levels is also constructed based on the risk factors. The risk factors for environmental complexity include weather type and traffic flow. The risk factors for task urgency include driving task type, minimum headway (THW), and minimum time to collision (TTC). The risk factors for interference intensity include stimulus mode and interference duration. The four risk levels are extremely low risk, low risk, medium risk, and high risk, respectively. Step S3: Setting up key test scenarios; The orthogonal experimental design method is used to determine the combination scheme of each risk factor in step S2, which needs to cover all four levels of risk scenarios. Specifically, the risk factors of environmental complexity, task urgency and interference intensity are orthogonally combined first. Then, the risk factor scenarios after orthogonal combination are selected, and four basic scenarios are selected according to risk weight. Finally, the basic scenarios of environmental complexity, task urgency and interference intensity are orthogonally combined again to form 4×4×4=64 basic scenarios. Step S4: Virtual simulation modeling; A joint simulation platform is used to construct virtual simulation scenarios and high-fidelity dynamic models. Scenarios involving risk factors are selectively added by the orthogonal experiments in step S3. Step S5: Real-world scenario reproduction; By modifying the actual vehicle site, laying a controllable water / snowmaking system and setting dynamic obstacles, a one-to-one mapping between the simulation scene and the actual vehicle scene can be achieved. Step S6: Multi-dimensional performance evaluation indicator system; A multi-dimensional performance evaluation index system with a three-layer structure of target layer, overall index layer, and index layer is constructed: the target layer is for multi-dimensional performance evaluation of the brain-controlled vehicle system, the overall index layer includes six dimensions: driving safety, anti-interference, generalization, timeliness, comfort, and data security, and the index layer consists of specific quantitative evaluation indicators under each overall index. The evaluation indicators for driving safety are TTC, THW, lane centerline deviation (LCO), and emergency braking distance (EBD); the evaluation indicators for anti-interference are function recognition accuracy (ERA), false trigger rate (FPR), and signal-to-noise ratio (SNR); the evaluation indicators for generalization are cross-subject accuracy (CPA) and cross-scenario adaptation rate (CSA); the evaluation indicators for timeliness are command response time (RT), information transmission rate (ITR), and command execution delay (ED); the evaluation indicators for comfort are mental workload (ML), device wearing comfort (WC), and operation satisfaction (OS); and the evaluation indicators for data security are data encryption strength (ES), data transmission packet loss rate (PLR), privacy data anonymization rate (PDR), and data storage integrity (ESI). The calculation method for TTC, one of the evaluation indicators for driving safety, is as follows: ； In the formula, For relative distance, The absolute value is taken when the relative velocity is negative. The absolute speed of a vehicle in a continuous series of vehicles; The speed of the vehicle targeting the obstacle; The calculation method is as follows: ； In the formula, The distance between the front ends of vehicles traveling in a continuous line; The calculation method for EBD is as follows: ； In the formula, For system response time, For braking and deceleration, This is the initial braking speed; The LCO is calculated as follows: ； In the formula, The lateral coordinates of the vehicle's center of gravity, acquired via an on-vehicle IMU or LiDAR, are shown perpendicular to the direction of travel. (Unit: [unit not specified]) ; The lateral coordinates of the lane centerline obtained through SCANeR scene modeling or high-precision real-vehicle maps, in units of ; Step S7: Construction of a closed-loop virtual-real testing platform integrating virtual simulation, real vehicle testing, data synchronization, and data security; The virtual simulation platform generates four levels of risk test scenarios: extremely low risk, low risk, medium risk, and high risk. It outputs scenario parameters and virtual vehicle status, and synchronizes scenario commands to the real vehicle subsystem. The subsystem consists of hardware and software systems. The real vehicle testing system is mounted on a modified experimental vehicle. The original driving control module of the experimental vehicle has been removed, and a brain control interface module has been added, which is compatible with the CANFD bus. The real vehicle testing subsystem specifically includes an EEG acquisition module, a vehicle status acquisition module, and a stimulus generation module. The data synchronization system can achieve time alignment of multi-source data from virtual simulation, real vehicles, and EEG, specifically including synchronization protocols, synchronization nodes, and unified data format processing. Data security is deeply integrated with virtual simulation, real vehicle testing, and data synchronization subsystems to protect sensitive biometric information. Step S8: Establish standardized evaluation methods; In step S7, on the four-in-one virtual-real closed-loop testing platform, a multi-dimensional performance evaluation index system is used as the evaluation standard to select suitable experimental groups for training, construct experimental scenarios, execute standardized experiments, acquire and analyze relevant data, and finally complete the establishment of standardized evaluation methods.

2. The method for performance evaluation and assessment of brain-controlled vehicle systems oriented towards key testing scenarios as described in claim 1, characterized in that, The specific implementation of the Savitsky-Golay filtering algorithm in step S1 is as follows: ； In the formula, To smooth and denoise the data, Here are the original data, where 'a' is the filter radius and 2a+1 is the sliding port length. This is the weighting factor.

3. The method for performance evaluation and assessment of brain-controlled vehicle systems oriented towards key testing scenarios as described in claim 1, characterized in that: In step S2, the risk factor classification of environmental complexity based on weather type is defined as follows: clear weather with visibility ≥ 20km is defined as extremely low risk; cloudy weather with visibility 10km-20km is defined as low risk; rainy weather with visibility 5km-10km is defined as medium risk; and foggy and snowy weather with visibility ≤ 5km is defined as high risk. The risk level classification based on traffic flow is defined as follows: scenarios with traffic flow ≤ 10 vehicles / km are defined as extremely low risk; scenarios with traffic flow 10-20 vehicles / km are defined as low risk; scenarios with traffic flow 20-30 vehicles / km are defined as medium risk; and scenarios with traffic flow > 30 vehicles / km are defined as high risk. The risk factor driving task type risk level division of task urgency is defined as: the straight-line constant-speed driving scenario is defined as a very low risk level, the smooth turning or following driving scenario is defined as a low risk level, the lane change and obstacle avoidance behavior scenarios are defined as medium risk levels, and the emergency braking and collision warning scenarios are defined as high risk levels; The THW risk level division is defined as: scenarios with THW≥5s are defined as very low risk levels, scenarios with THW 3s - 5s are defined as low risk levels, scenarios with THW 2s - 3s are defined as medium risk levels, and scenarios with THW≤2s are defined as high risk levels; The TTC risk level division is defined as: scenarios with TTC≥8s are defined as very low risk levels, scenarios with TTC 5s - 8s are defined as low risk levels, scenarios with TTC 3s - ​ ​ 4. The method for performance evaluation and assessment of brain-controlled vehicle systems oriented towards key testing scenarios as described in claim 1, characterized in that: ​ 5. The method for performance evaluation and assessment of brain-controlled vehicle systems oriented towards key testing scenarios according to claim 1, characterized in that: ​ 6. The method for performance evaluation and assessment of brain-controlled vehicle systems oriented towards key testing scenarios according to claim 1, characterized in that: ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ Anti-interference capability: The calculation method is as follows: ； In the formula, To correctly identify the number of times, This represents the number of incorrect identifications. ​ The calculation method for FPR is as follows: ； In the formula, Number of false positives The number of true negatives; Definition: Under normal driving conditions on highways in clear weather, the FPR safety threshold is ≤0.1%; under urban road conditions, the FPR safety threshold is relaxed to ≤0.5%. The method for calculating SNR is as follows: ； In the formula, This represents the effective value of the signal amplitude. The effective value representing the amplitude of the signal noise. The amplitude of the alpha wave in the brain. This represents the amplitude of electromyographic interference. Definition: Under daytime heavy rain conditions on highways, the SNR safety threshold is ≥15dB; under ordinary urban road conditions, the SNR safety threshold is ≥10dB. Generalization: The calculation method for CPA is as follows, ； In the formula, For the number of subjects, For the first The accuracy of the number of subjects; Definition: Basic pass threshold CPA ≥ 80%; Excellent threshold CPA ≥ 85%, which means meeting the multi-user scenario requirements of L2+ level assisted driving system; The calculation method for CSA is as follows: ； In the formula, For the number of scenes, For the scene Number of times the function was successfully implemented in China This represents the total number of tests. Definitions: Low-difference scenario, from sunny to cloudy urban areas, CSA ≥ 90%; Medium-difference scenario, from sunny to rainy urban areas, and from urban areas with streetlights at night, CSA ≥ 80% for rainy days and CSA ≥ 85% for nighttime; High-difference scenario, from sunny to rainy urban areas without streetlights, CSA ≥ 70%. Timeliness: The calculation method for RT is as follows, ； In the formula, To record the timing of EEG intention triggering based on the TSN time synchronization system, This refers to the timing of command execution by the vehicle ECU based on the TSN time synchronization system. Definition: In high-speed cruising scenarios, the normal response time (RT) is ≤0.5s, and in emergency scenarios, the RT is ≤0.3s; in urban road scenarios, the RT is ≤0.8s. The calculation method for ITR is as follows: ； In the formula, For single selection of time, For the number of optional targets, For accuracy; Definition: In the context of highways, the basic ITR is ≥8 bits / s, the total transmission time of critical information is ≤0.3s, and the proportion of redundant information is ≤10%. The calculation method for ED is as follows: ； In the formula, Record the CANFD bus command transmission time. The moment when the vehicle actuator operates; Definitions: Under normal high-speed operation, ED ≤ 0.5s; under high-speed emergency avoidance, ED ≤ 0.3s; under low-speed urban scenarios, ED ≤ 1s. Comfort: The calculation method for ML is as follows, ； In the formula, For dimension weights, The dimension score (0-100 points) is where n represents the core dimension of ML. The threshold for mental workload (ML) needs to be set in conjunction with the driving scenario. In a normal driving scenario, ML ≤ 3 points is considered too low load, corresponding to a driving state of inattention and slow reaction. ML 3-7 points is considered a comfortable load, corresponding to a driving state of focused attention and easy cognition. ML > 7 points is considered too high load, corresponding to a driving state of cognitive fatigue and decision-making errors. The NASA-TLX scale is used to calculate the weighted average score of six dimensions, including mental workload, physical workload, time workload, effort level, frustration, and performance. The calculation method for WC is as follows: ； In the formula: The overall score for the comfort of wearing the device is given, ranging from 0 to 1. The closer the score is to 1, the higher the comfort level. A 9-point comfort scale was used to rate driver comfort. This is an objective physiological indicator, expressed as local pressure, in kPa. This is an objective physiological indicator, expressed as changes in skin moisture, with units of g / m². It is an objective physiological indicator, expressed as muscle tension, with the unit being μV; The OS is calculated as follows: ； In the formula, OS is the total operation satisfaction score, which ranges from 0 to 100. The weight of the i-th evaluation dimension is set according to the importance of the scenario, and the sum of the weights of all dimensions is 1. The score for the i-th evaluation dimension ranges from 0 to 10; n is the number of evaluation dimensions. Data security: The calculation method is as follows: ； In the formula, The final data encryption strength score has a range of values. A score of 80 or above is considered Scene safety compliance standards; , , Weight coefficients for each dimension, satisfying The weights need to be based on Specific sub-scenes are dynamically adjusted. The encryption algorithm is rated on its anti-cracking complexity [0, 100] to measure its ability to resist brute-force and differential attacks; The security score for the entire lifecycle management of keys is [0, 100], which measures the protection capabilities of key generation, storage, update, and destruction. The adaptability score for data transmission / storage scenarios is [0, 100]. In urban road vehicle-to-vehicle communication (V2V) conditions, the security threshold for ES (Electronic Security) is ≥7 points. The specific classifications and corresponding risks are as follows: An ES score of 8-10 points indicates a high security level, with extremely strong resistance to hacking and the ability to withstand national-level computing power attacks. It ensures data authenticity and integrity, and there are no security risks in system decision-making. An ES score of 7-8 points represents the security threshold, capable of resisting conventional hacker attacks, ensuring data transmission and storage security; only extreme computing power attacks could lead to risks. An ES score of 5-7 points indicates a low security level, easily cracked by professional hackers, potentially leading to data tampering and increased risk of collision warning delays. An ES score <5 points indicates an insecure level, with no effective encryption or easily cracked encryption, making data highly susceptible to tampering. This could cause the system to misjudge distances, miss collision warnings, and lead to accidents. The calculation method for PLR is as follows: ； In the formula, For the number of lost packets, Total number of data packets; In urban V2X conditions, there is high vehicle density, frequent lane changes, and many unexpected scenarios. When the PLR ​​is in the range of 0~0.5%, it is at the highest safety level. There is almost no packet loss in data transmission, the calculation error is <1%, and the early warning response delay is <10ms, which fully meets the safety decision-making requirements. When PLR is in the range of 0.5% to 2%, it is considered a safe level. Occasional packet loss can be compensated for by the data retransmission mechanism. The calculation error is <3% and the warning delay is <50ms, with no security risk. When PLR is in the range of 2% to 5%, it is considered a critical risk level. The frequency of packet loss increases and the retransmission mechanism cannot fully cover the impact. The calculation error reaches 5% to 8% and the warning delay exceeds 100ms, requiring a speed reduction warning to be triggered. When PLR is >5%, it is considered a high-risk level. A large amount of data loss will lead to calculation distortion and the failure of the warning, directly increasing the risk of collision and requiring the triggering of the emergency avoidance mode. The calculation method for PDR is as follows: ； In the formula, This refers to the amount of sensitive data that has been anonymized. Total amount of sensitive data; Intelligent driving on urban roads is a scenario with high privacy risks, requiring a PDR (Performance Recognition Disclosure) rate of ≥95%, with a 100% anonymization rate for key features involved in identity recognition. The calculation method for DSI is as follows: ； In the formula, Total amount of data stored To determine the total amount of data to be stored, the DSI value range is: ;when When the value is 1, it means that all stored data has passed the integrity check and there is no damage; when the value is 1, it means that all stored data has passed the integrity check and there is no damage. When this occurs, it means that all stored data is corrupted or lost, resulting in a complete loss of integrity. Definition: When DSI ≥ 99.9%, it is a safe level, the core data is extremely complete, the calculation is accurate, and there is no safety risk; when DSI is in the range of 99.0% to 99.9%, it is a warning level, a small amount of non-core data is missing or damaged, resulting in a slight deviation in calculation accuracy; when DSI < 99.0%, it is a dangerous level, the core data is lost or tampered with, and it is necessary to urgently downgrade to manual driving.

7. The method for performance evaluation and assessment of brain-controlled vehicle systems oriented towards key testing scenarios according to claim 1, characterized in that: The virtual simulation platform in step S7 adopts a joint simulation platform of ScaneR Studio 2025 and Prescan. The hardware system of its subsystem specifically includes an industrial-grade server, an Intel Xeon Gold 6426R CPU, an NVIDIA A100 GPU, 256GB DDR5 memory, and a 4K three-screen projection system; the software system specifically includes the scene modeling software ScaneR Studio 2025, the vehicle dynamics simulation software Prescan 2024, and the EEG signal simulation software MATLAB / Simulink. The EEG acquisition module of the real vehicle test system is equipped with a Neuroscan SynAmps RT multi-channel EEG instrument: 64 channels, 1000Hz sampling rate, input impedance ≥10GΩ, and Ag / AgCl electrode device to collect information from key brain regions, with specific sampling points covering key brain regions. The vehicle status acquisition module is equipped with millimeter-wave radar: detection range 0.1-200m, accuracy ±0.1m; lidar: 128 lines, point cloud density 200 points / ㎡; inertial measurement unit (IMU): sampling rate 1000Hz, acceleration accuracy ±0.01m / s². The stimulation generation module is equipped with an in-vehicle display screen for visual cues, an in-vehicle audio system for auditory cues (frequency 500-1000Hz), and a seat vibration module for tactile cues (vibration frequency 20-50Hz, amplitude 0.5-2mm). The data synchronization system uses a 5G+Time Sensitive Network (TSN) synchronization protocol. The synchronization node is achieved by deploying three time synchronization nodes: a scenario node for the simulation subsystem, an EEG node for the EEG acquisition module, and a vehicle node for the real vehicle subsystem. Each node is connected through a Time-Sensitive Network (TSN) switch. The data is uniformly encapsulated in JSON-LD format, including timestamp UTC time and local nanosecond offset; the data types specifically include scene, EEG, and vehicle; the data content includes EEG amplitude and vehicle speed.

8. The method for performance evaluation and assessment of a brain-controlled vehicle system oriented towards key testing scenarios according to claim 1, characterized in that, The preparation phase of the standardized evaluation method in step S8 is as follows: Selecting and training suitable candidates: Candidates must meet the following screening criteria: age 25-40, driving experience ≥3 years and holding a C1 or above driver's license, no brain or mental illness, normal vision or corrected vision ≥1.0, pass the Mini-Mental State Examination (MMSE) with a score ≥27, which indicates normal cognitive ability. The training includes a 2-hour pre-experiment session, specifically covering training on wearing the EEG device to ensure electrode impedance ≤5kΩ to guarantee the accuracy of EEG signal acquisition, brain control function training to enable subjects to operate it proficiently, and scenario adaptation training to familiarize subjects with the operational logic of scenarios with different risk levels. After the training, subjects should achieve a brain control function recognition accuracy rate of ≥80%. Experimental scenarios were constructed: Based on the risk level of the scenarios, the experiment was divided into four groups: extremely low risk group, low risk group, medium risk group, and high risk group. Each group contained five core scenarios. Each scenario tested five brain-controlled functions: raising and lowering car windows, adjusting air conditioning temperature, switching music, adjusting volume, and braking. Regarding speed control, the extremely low risk group is 15-30 km / h, the low risk group is 10-25 km / h, the medium risk group is 8-20 km / h, and the high risk group is 5-15 km / h; the stimulation type is divided into weak stimulation and strong stimulation. Each scenario and function combination was tested 10 times to ensure that the experimental data were statistically significant and that the sample size was ≥50.

9. The method for performance evaluation and assessment of a brain-controlled vehicle system oriented towards key testing scenarios according to claim 1, characterized in that, The standardized experiment for the standardized evaluation method in step S8 is as follows: Step S81: After the subject arrives at the laboratory, he / she first fills out an informed consent form, then puts on the EEG device, and at the same time performs electrode impedance calibration to ensure that the impedance of all channels is ≤5kΩ to ensure normal acquisition of EEG signals. Step S82: Start the test platform, synchronize the virtual simulation subsystem, the real vehicle subsystem and the data synchronization subsystem, and confirm that the communication between each module is normal; Step S83: Test in order of scenario risk level from low to high to avoid the test results of high-risk scenarios being affected by subject fatigue. Before each scenario begins, the simulation subsystem will output scenario prompts to let the subject know the scenario information. Step S84: During the scenario operation, the subject performs corresponding brain control functions according to the stimulus prompts. During this process, the data synchronization subsystem collects EEG data, vehicle data, and scenario data in real time.

10. The method for performance evaluation and assessment of a brain-controlled vehicle system oriented towards key testing scenarios according to claim 1, characterized in that, The data processing method for the standardized evaluation method in step S8 is as follows: Data processing consists of two parts: data preprocessing and data backup. In data preprocessing, the Savitsky-Golay filter with a window length of 5 and a polynomial order of 2 is used to remove jitter from vehicle data. For EEG data, an 8Hz high-pass filter + a 30Hz low-pass filter is used to remove 50Hz power frequency interference and electromyographic noise. In the data backup phase, experimental data is stored in real time on edge computing nodes equipped with encrypted hard drives, and uploaded to the cloud distributed storage Alibaba Cloud OSS within 24 hours. At the same time, a SHA-256 hash value is generated for data integrity verification to prevent data loss or tampering.

Images