Vehicle control method and vehicle control device

Abstract

The invention discloses a vehicle control method and device based on multi-source information fusion and actuator health state monitoring. The vehicle control method comprises the steps that driver state data, vehicle dynamics data and environment information data are obtained; sensing and fusing the multi-source data to generate comprehensive sensing data; performing scene classification according to the comprehensive perception data, identifying and judging a current driving scene type, and determining a control demand of current driving; health state monitoring is conducted on vehicle actuators, and the working performance and reliability of all the actuators are evaluated in real time; according to a scene judgment result and an actuator health state monitoring result, dynamically adjusting weight distribution of each control target in a decision making process; and based on the adjusted weight distribution, executing multi-target collaborative optimization, generating an optimal control strategy, converting the optimal control strategy into a control instruction, and issuing the control instruction to a vehicle execution mechanism. And the technical bottlenecks of the existing system in complex scene adaptability, actuator collaborative optimization, fault handling capacity and the like can be solved.

Description

Technical Field

[0001] This invention relates to the field of vehicle control technology, and in particular to a vehicle control method and a vehicle control device. Background Technology

[0002] With the rapid development of intelligent driving technology, distributed drive vehicle platforms have gradually become a research hotspot. These platforms achieve efficient collaboration among multiple execution systems in the chassis domain through the coordinated control of drive, braking, steering, and suspension systems. However, designing a scenario-adaptive multi-execution system collaborative scheduling mechanism to fully exploit the performance potential of the distributed drive platform has become a core challenge that urgently needs to be overcome in the field of intelligent chassis domain control technology. Summary of the Invention

[0003] This invention aims to address at least one of the technical problems existing in the prior art. To this end, this invention proposes a vehicle control method that can overcome key technical bottlenecks in existing systems regarding adaptability to complex scenarios, actuator collaborative optimization, and fault response capabilities.

[0004] The present invention further proposes a vehicle control device.

[0005] According to a first aspect of the present invention, a vehicle control method based on multi-source information fusion and actuator health status monitoring includes the following steps: acquiring driver state data, vehicle dynamics data, and environmental information data; performing multi-source information perception and fusion on the driver state data, vehicle dynamics data, and environmental information data to generate fused comprehensive perception data; classifying scenarios based on the comprehensive perception data to identify the current driving scenario type; determining the control requirements under the current driving situation based on the scenario classification results; monitoring the health status of vehicle actuators to evaluate the working performance and reliability of each actuator in real time; dynamically adjusting the weight allocation of each control objective in the decision-making process based on the scenario judgment results and actuator health status monitoring results; performing multi-objective collaborative optimization based on the adjusted weight allocation to generate an optimal control strategy; and converting the optimal control strategy into control commands and issuing them to the vehicle actuators to realize the execution of the control commands.

[0006] Therefore, this invention constructs a complete intelligent driving vehicle control method. First, it employs multimodal fusion technology to collect and fuse driver state data, environmental information data, and vehicle state data in real time, classifying driving scenarios into different types. Second, it establishes a dynamic weight allocation mechanism, combining scenario classification results with real-time system state parameters. This incorporates cross-system redundancy decision-making mechanisms into a multi-objective collaborative optimization process, dynamically adjusting the control weights of each actuator through adaptive algorithms. This achieves functional complementarity and load optimization among the drive, braking, steering, and suspension systems, forming a complete technical closed loop of perception-decision-execution. This significantly improves the system's reliability, adaptability, and energy efficiency under complex operating conditions, providing crucial technical support for high-level autonomous vehicles. This invention solves key technical bottlenecks in existing systems regarding adaptability to complex scenarios, actuator collaborative optimization, and fault response capabilities, providing a safer and more efficient chassis control solution for high-level autonomous vehicles.

[0007] According to some embodiments of the present invention, the driver state data is a driver state index ( ), For a scalar value between [0, 1], it is calculated by the following formula:

[0008] in, For fatigue component, To distract from the task, Let α, β, and γ be the stress components; α, β, and γ be the weighting coefficients, satisfying... + + =1; The driving risk level is dynamically divided based on the numerical range of the driver state index.

[0009] According to some embodiments of the present invention, the vehicle dynamics data is the vehicle stability margin ( The result is obtained by the following formula:

[0010] in, This represents the absolute value of the vehicle's actual lateral acceleration. The current road surface adhesion coefficient, This represents the theoretical maximum lateral acceleration of the current road surface. g is the acceleration due to gravity; the vehicle stability state level is dynamically divided based on the numerical range of the vehicle stability margin. According to some embodiments of the present invention, the environmental information data is the environmental threat level (…). The result is obtained by the following formula:

[0011] in, The minimum radius of curvature of the road ahead; The time of collision with the nearest obstacle in front; The current lane width; The road surface adhesion coefficient; , , , Let be the weighting coefficient, satisfying The environmental threat level is dynamically divided based on the numerical range of the environmental threat level using the environmental information data.

[0012] According to some embodiments of the present invention, the fatigue component Based on eye-tracking metrics, the expression is:

[0013] in, The percentage of time within 60 seconds during which the eyelids are closed at more than 80% of the time. Blink frequency (unit: times / minute). 2 is the weighting coefficient, which satisfies 1 and .

[0014] According to some embodiments of the present invention, the distraction component Based on the calculation of gaze deviation and head posture, the expression is: ) in, It is a statistic of the time the viewpoint leaves the single-path area in front within 60 seconds. The threshold time for the line of sight to leave the area of ​​the road ahead; This refers to the angle at which the head deviates from the forward direction. It is the threshold angle at which the head deviates from the forward direction.

[0015] According to some embodiments of the present invention, the stress component Based on heart rate variability (HRV), the expression is:

[0016] It is the ratio of its low-frequency power to its high-frequency power.

[0017] According to some embodiments of the present invention, the step of classifying scenarios and identifying the current driving scenario type based on the comprehensive perception data includes: inputting the comprehensive perception data into a neural network model for a first-level risk assessment, outputting a preliminary risk level, including low risk, medium risk, high risk, or passability risk; when the preliminary risk level is high risk, performing a second-level risk assessment, further subdividing the high risk into risk levels based on the dominant risk factors; wherein, if the risk mainly stems from the vehicle's dynamic state approaching its limit, it is determined to be a stability risk; if the risk mainly stems from the driver's lack of control ability while the vehicle still has stability, it is determined to be a handling risk; finally, different driving scenario types are output according to different risk results; wherein, the driving scenario includes at least: "normal driving", "on the verge of instability", "instability", and "getting out of trouble".

[0018] According to some embodiments of the present invention, the step of determining the control requirements under the current driving situation based on the scenario classification result includes: the scenario classification result is sent to the chassis domain controller, the chassis domain controller calls the preset control strategy corresponding to the driving scenario type and executes vehicle control operation according to the received scenario classification result; at the same time, the performance evaluation module starts working, and monitors the performance indicators characterizing the current scenario classification result and its corresponding system control effect in real time; when the performance indicators meet the preset excellent conditions, the existing scenario classification rules and thresholds are maintained unchanged; when the performance indicators continuously deviate from the expected range, or the scenario classification result changes frequently with high control intervention intensity, an online learning mechanism is triggered; the online learning mechanism collects a data sequence containing comprehensive perception data, scenario classification results, control commands and performance indicators, and generates adjustment amounts for classification rules and thresholds through a lightweight incremental learning algorithm; the adjustment amounts are used to make minor online updates to the decision parameters of the scenario type within a preset safety constraint range, so that the updated classification rules and thresholds are used immediately for subsequent scenario judgment, thereby forming a closed-loop adaptive optimization.

[0019] According to some embodiments of the present invention, after the step of determining the control requirements under the current driving situation based on the scenario classification results, the system further includes: according to the current driving scenario type, the system dynamically configures the priority order of multiple control performance indicators based on the safety requirements and control objectives of the scenario, and adjusts the weight allocation in the multi-objective collaborative optimization process based on the priority order; the multi-objective collaborative optimization is constructed based on the multi-dimensional deviation between the current state of the vehicle and the desired target state, the multi-dimensional deviation including at least path tracking deviation, vehicle stability deviation, and ride comfort-related deviation; the weight allocation of various deviations in the optimization objectives is dynamically adjusted according to the driving scenario, and the control command sequence with the minimum comprehensive cost is output.

[0020] According to some embodiments of the present invention, the system dynamically configures the priority order of multiple control performance indicators based on the safety requirements and control objectives of the scenario, and adjusts the weight allocation in the multi-objective collaborative optimization process based on the priority order, including: constructing the control objective as an optimization objective function ( Its form is:

[0021] in, The state error vector includes at least tracking error, stability error, and comfort error. To control the input quantity, it includes all vehicle actuator actions, which at least cover vehicle actuator commands for the braking, steering, drive and suspension systems; This is the state weight matrix; It is the control weight matrix; the solution is to minimize J.

[0022] According to some embodiments of the present invention, the system dynamically assigns priority to various control performance indicators based on the current driving scenario type, and quantifies each control performance indicator into an initial weight vector. It is expressed by the following formula: .

[0023] in, The initial weight represents the i-th control objective, and S represents the driving scenario identifier. The system generates the corresponding initial state weight and control weight based on the current driving scenario identifier through a preset scenario-weight mapping relationship. The weight is used to construct the objective function of the multi-objective optimization problem. The system solves the optimal control command sequence online based on the dynamic weight and sends the first element of the command to each of the vehicle actuators to realize multi-system collaborative control.

[0024] According to some embodiments of the present invention, the step of dynamically adjusting the weight allocation of each control objective in the decision-making process based on the scenario judgment result and the actuator health status monitoring result includes: dynamically allocating the priority order of multiple control performance indicators according to the current driving scenario type, and quantifying each control performance indicator into an initial weight vector. When any actuator is detected to have experienced performance degradation or failure, the system proportionally reduces the initial weight of that actuator in the control input vector and renormalizes the weights of the remaining healthy actuators, so that the control tasks originally undertaken by the failed actuator are automatically assigned to the functionally relevant healthy actuators; the multi-objective optimization problem after weight adjustment aims to minimize the overall cost, and generates and issues an updated sequence of control instructions.

[0025] According to some embodiments of the present invention, when any actuator is detected to have experienced performance degradation or failure, the system proportionally decays the initial weight vector corresponding to that actuator in the control input vector and renormalizes the weights of the remaining healthy actuators, so that the control tasks originally undertaken by the failed actuator are automatically allocated to the functionally relevant healthy actuators, including: receiving the initial weight vector. A health state vector H is generated, where each element of the health state vector represents the functional integrity of the corresponding actuator. It is then determined whether there is a non-1 value in the health state vector H. If not, the original weights remain unchanged. If so, the weight components corresponding to each actuator with a health state less than 1 are proportionally attenuated, with the attenuated weight component equal to the original weight multiplied by the corresponding health state value. All attenuated weight components are aggregated to form a downgraded weight vector, which is then normalized to generate an updated weight vector. This updated weight vector is used as input for subsequent optimization calculations to adjust the weight allocation.

[0026] According to some embodiments of the present invention, the vehicle control method further includes: real-time monitoring of the health status vector of each actuator and transmitting it to the control system via the vehicle bus; when a continuous decay of the health status value of a certain actuator is detected, the control output of the actuator is reduced proportionally and smoothly according to the change of its health status value; and the control output of the redundant actuators that need to be added is calculated synchronously to compensate for the reduced control output of the decaying actuators, so as to ensure that the overall performance indicators of the vehicle remain unchanged or fluctuate within a preset range.

[0027] According to a second aspect of the present invention, a vehicle control device includes: a data acquisition unit for real-time acquisition of driver status data, vehicle dynamics status data, and environmental information data; an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the vehicle control method when running the computer program; an execution unit communicatively connected to the electronic device for receiving control commands from the electronic device and executing corresponding vehicle control actions, wherein the execution unit includes at least a plurality of actuators in a drive system, a braking system, a steering system, and a suspension system; and a communication interface integrated into the electronic device for enabling data interaction between the electronic device and the data acquisition unit, the execution unit, and other on-board or off-board devices.

[0028] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0029] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a general flowchart of a vehicle control method according to an embodiment of the present invention; Figure 2 This is a logical diagram illustrating driving scenario classification according to an embodiment of the present invention; Figure 3 This is a logical diagram illustrating the dynamic updating of scene classification rules according to an embodiment of the present invention; Figure 4 This is a logical diagram of dynamic weight allocation according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0030] The embodiments of the present invention are described in detail below. The embodiments described with reference to the accompanying drawings are exemplary. The embodiments of the present invention are described in detail below.

[0031] The following is for reference. Figures 1-5 This invention describes a vehicle control method and a vehicle control device based on multi-source information fusion and actuator health status monitoring according to embodiments of the present invention.

[0032] The vehicle control method based on multi-source information fusion and actuator health status monitoring according to a first aspect of the present invention includes the following steps: Acquire driver status data, vehicle dynamics data, and environmental information data; Multi-source information perception and fusion of driver status data, vehicle dynamics data and environmental information data are performed to generate integrated perception data after fusion. Based on comprehensive perception data, the current driving scenario type is identified through scenario classification. Based on the scenario classification results, scenario judgment is performed to determine the control requirements in the current driving situation; Health status monitoring of vehicle actuators, and real-time evaluation of the working performance and reliability of each actuator; Based on the scenario judgment results and actuator health status monitoring results, the weight allocation of each control objective in the decision-making process is dynamically adjusted; Based on the adjusted weight allocation, multi-objective collaborative optimization is performed to generate the optimal control strategy; The optimal control strategy is transformed into control commands and sent to the vehicle actuators to execute the control commands.

[0033] like Figure 1As shown, multimodal fusion technology is first employed to collect and fuse driver status data, environmental information, and vehicle status information in real time through onboard sensors and external sensing devices. The required data includes: driver physiological status data (including but not limited to eye movement, heart rate, body temperature, pulse, and respiration) collected through in-vehicle cameras, eye trackers, and physiological sensors; vehicle status information (including but not limited to yaw rate, vehicle mass, center of gravity sideslip angle, steering angle, vehicle speed, and tire sideslip stiffness) collected through vehicle sensors; and environmental data (such as lane markings, road slope, road surface adhesion coefficient, obstacles, pedestrians, traffic signs, and weather conditions) acquired through cameras, LiDAR, millimeter-wave radar, and GPS. Subsequently, relevant algorithms or models are applied to the above multi-source information data to achieve multi-level information fusion, generating comprehensive perception data that fully reflects the current driving situation.

[0034] Based on this, the system analyzes the fused comprehensive perception data according to preset scenario classification rules or trained classification models to identify the current driving scenario type. For example, it can be divided into "low risk," "medium risk," "high risk," or "passability risk"; or it can be divided into specific scenarios such as highway cruising, urban intersection driving, emergency obstacle avoidance, and driving on slippery roads. Of course, driving scenario types include, but are not limited to, the above-mentioned embodiments.

[0035] Next, based on the identified scenario type, the system further determines the priority of control requirements within that scenario. For example, in normal driving scenarios, the control objective is to provide a smooth driving experience with minimal intervention, prioritizing driving comfort / economy over handling and safety. In near-instability scenarios, the control objective is to actively and smoothly assist the driver in restoring stability and preventing instability, prioritizing safety (pre-stability) over handling and comfort. In instability scenarios, the control objective will take all necessary measures to quickly and effectively correct the vehicle's posture and stabilize it, prioritizing safety (stability) over comfort. In obstacle-avoidance scenarios, the control objective will maximize traction to assist the vehicle in overcoming obstacles, prioritizing passability over safety and comfort. Furthermore, in specific scenarios, lateral stability and collision avoidance capabilities are prioritized in emergency obstacle avoidance scenarios, while fuel economy and ride comfort are emphasized in high-speed cruising scenarios.

[0036] Meanwhile, the system continuously monitors the operating status of various actuators in the vehicle (such as the electronic stability control system, steering system, braking system, suspension system, drive motor, etc.), and assesses their health status and reliability level by collecting actuator feedback signals, response delays, output deviations, and other indicators in real time. If an actuator experiences performance degradation or potential failure, the system will reduce its weight in the control allocation or even isolate it from the control loop.

[0037] Based on the scenario judgment results and actuator health status assessment results, the system dynamically adjusts the weight allocation of multiple control objectives (such as safety, stability, comfort, and energy efficiency) in the decision optimization process. For example, when a part of the actuator fails, the system will significantly increase the safety weight and moderately sacrifice comfort to ensure basic controllability.

[0038] Subsequently, the system constructs a multi-objective optimization problem using the adjusted weights as constraints. The optimal control strategy can be solved using model predictive control, quadratic programming, or other advanced control algorithms. This strategy comprehensively considers vehicle dynamics, current scenario constraints, and actuator capability boundaries to ensure the feasibility and robustness of control commands.

[0039] Ultimately, the system transforms the generated optimal control strategy into specific control commands (such as target braking force distribution, desired yaw moment, steering angle commands, etc.), and sends them to the corresponding actuators through the vehicle network to complete closed-loop control, thereby achieving safe, efficient, and reliable coordinated control of the vehicle.

[0040] Based on the aforementioned control methods, existing technologies for multimodal scene recognition primarily rely on vehicle state information such as yaw rate and wheel speed difference, employing fixed threshold rules for scene judgment, which easily leads to misjudgment. This invention, however, integrates driver feature information, environmental information, and vehicle state information, achieving human-vehicle-environment collaborative perception through a dynamic scene classifier. This breaks through the traditional sequential judgment logic based on a single vehicle state, constructing a multi-dimensional collaborative parallel processing mechanism. This difference improves the system's scene recognition accuracy in complex conditions such as off-road driving, effectively avoiding false triggering issues such as misinterpreting a driver's normal observation of obstacles as a need for escape.

[0041] Regarding control strategies, existing technologies employ fixed parameter mapping relationships. For example, the "getting out of trouble" mode in existing technologies typically involves uniformly increasing torque output without considering real-time monitoring of actuator health status. This invention proposes a dynamic adaptive control algorithm that establishes a scene type-actuator dynamic coupling model. It correlates the scene classifier output with actuator health status indicators in real time and incorporates the execution system health status information into a multi-objective collaborative optimization weight allocation strategy. This achieves the optimal scheduling problem in a given task process, realizing a leap from information layer redundancy to decision-making and functional layer redundancy.

[0042] Therefore, this invention constructs a complete intelligent driving vehicle control method. First, it employs multimodal fusion technology to collect and fuse driver state data, environmental information data, and vehicle state data in real time, classifying driving scenarios into different types. Second, it establishes a dynamic weight allocation mechanism, combining scenario classification results with real-time system state parameters. This incorporates cross-system redundancy decision-making mechanisms into a multi-objective collaborative optimization process, dynamically adjusting the control weights of each actuator through adaptive algorithms. This achieves functional complementarity and load optimization among the drive, braking, steering, and suspension systems, forming a complete technical closed loop of perception-decision-execution. This significantly improves the system's reliability, adaptability, and energy efficiency under complex operating conditions, providing crucial technical support for high-level autonomous vehicles. This invention solves key technical bottlenecks in existing systems regarding adaptability to complex scenarios, actuator collaborative optimization, and fault response capabilities, providing a safer and more efficient chassis control solution for high-level autonomous vehicles.

[0043] In some embodiments of the present invention, the driver state data is the driver state index ( ), For a scalar value between [0, 1], it is calculated by the following formula:

[0044] in, For fatigue component, To distract from the task, Let α, β, and γ be the stress components; α, β, and γ be the weighting coefficients, satisfying... + + =1; Dynamically classify driving risk levels based on the numerical range of the driver's state index.

[0045] Understandably, the first step involves the quantitative calculation of the driver state index, a scalar value between [0, 1] used to comprehensively quantify the driver's physiological arousal and cognitive load levels. Its calculation is based on a multi-feature weighted fusion model: Where α, β, and γ are weighting coefficients, and the allocation of weighting coefficients is based on the instantaneousness, directness, and severity of the driving safety risk to different states.

[0046] Given that distraction can directly lead to a collision in a very short time, it has the highest risk priority; fatigue cumulatively reduces driving ability, so its risk priority is second; the impact of stress on driving is relatively indirect and complex, so its risk priority is third. Therefore, the weighting coefficients preferably satisfy the relationship β > α > γ. Specific values ​​can be determined by training a labeled dataset using supervised machine learning methods, or by calibration based on a large amount of real-vehicle test data using expert experience. In a preferred embodiment, the weighting coefficients are taken in the following ranges: distraction weighting coefficient β: 0.4~0.5; fatigue weighting coefficient α: 0.3~0.4; stress weighting coefficient γ: 0.1~0.2.

[0047] In some embodiments of the present invention, vehicle dynamics data are vehicle stability margins ( The result is obtained by the following formula:

[0048] in, This represents the absolute value of the vehicle's actual lateral acceleration. The current road surface adhesion coefficient, This represents the theoretical maximum lateral acceleration of the current road surface. , where g is the acceleration due to gravity; the vehicle stability state level is dynamically divided based on the numerical range of the vehicle stability margin.

[0049] Specifically, The absolute value of the vehicle's actual lateral acceleration can be obtained in real time through the onboard inertial measurement unit (IMU). This is the current road surface adhesion coefficient estimated by the system in real time. This represents the theoretical maximum lateral acceleration on the current road surface. Road surface adhesion coefficient. It can be jointly estimated by integrating the historical excitation response of the vehicle's ESP system, the recognition results of road surface material texture by the visual camera, and the weather information broadcast by the network.

[0050] Based on the specific data of VSM, it can be divided into: VSM≈1: Indicates that the vehicle is in a completely stable state, the lateral acceleration is far below the limit, and there is sufficient stability margin. VSM≈0: Indicates that the actual lateral acceleration of the vehicle is close to the physical limit that the current road surface can provide. → When the stability margin is exhausted, the vehicle is on the verge of instability. VSM<0: This indicates that the vehicle has become unstable, the tire force is saturated, and the lateral acceleration exceeds the theoretical limit (possibly due to measurement delay or estimation error).

[0051] Furthermore, the VSM threshold is dynamically set according to control requirements: when the VSM is below the first threshold (preferred value 0.35), the vehicle stability margin is considered low; when the VSM is below the second threshold (preferred value 0.15), the vehicle is considered to be on the verge of or already in an unstable state. It can be understood that the numerical range [0, 1] has been further refined. First, it is determined whether the VSM is below the first threshold; if so, it indicates that the vehicle stability is low. Then, it is further determined whether the VSM is below the second threshold; if so, it indicates that the vehicle is on the verge of or already in an unstable state, which is relatively dangerous.

[0052] In some embodiments of the present invention, environmental information data is the environmental threat level (…). The result is obtained by the following formula:

[0053] in, The minimum radius of curvature of the road ahead; The time of collision with the nearest obstacle in front; The current lane width; The road surface adhesion coefficient; , , , Let be the weighting coefficient, satisfying Environmental threat levels are dynamically classified based on the numerical range of environmental threat levels using environmental information data.

[0054] This can be obtained from a forward-facing camera or a high-precision map. The statement indicates that the sharper the bend, the greater the threat. It can be measured by radar or camera. The shorter the collision time, the greater the threat. The term indicates that the narrower the lane, the smaller the margin for error, and the greater the threat. 1 / The item indicates that the more slippery the road surface, the greater the threat.

[0055] In the above formula, each weight coefficient is allocated based on its degree of impact on safety, prioritizing the satisfaction of... > > > The relationship (the sum of all weight coefficients is 1). The threshold can be determined by a machine learning classifier based on a large amount of scene data. This involves inputting the environmental feature vector into a pre-trained support vector machine (SVM) classification model, which can output three environmental threat levels: low, medium, and high.

[0056] In some embodiments of the present invention, fatigue component Based on eye-tracking metrics, the expression is:

[0057] in, The percentage of time within 60 seconds during which the eyelids are closed at more than 80% of the time. Blink frequency (unit: times / minute). 2 is the weighting coefficient, which satisfies 1 and .

[0058] It is understandable that when the eyelid closure degree exceeds 80%, it indicates that the driver is in a state of severe fatigue. Therefore, the percentage of time within 60 seconds when the eyelid closure degree exceeds 80% is used as the core indicator, while the blinking frequency is used as a secondary indicator.

[0059] Preferably, based on Higher reliability of indicators in fatigue assessment The value takes the range [0.6, 0.8]. The value is in the range of [0.2, 0.4], and the specific value can be obtained through experimental data calibration or machine learning training.

[0060] In some embodiments of the present invention, distraction component Based on the calculation of gaze deviation and head posture, the expression is: ) in, It is a statistic of the time the viewpoint leaves the single-path area in front within 60 seconds. The threshold time for the line of sight to leave the road area in front is preferably 2 seconds. This refers to the angle at which the head deviates from the forward direction. It is the threshold angle at which the head deviates from the forward direction. Preferably, the threshold angle is 20°.

[0061] In some embodiments of the present invention, the stress component Based on heart rate variability (HRV), the expression is:

[0062] It is the ratio of its low-frequency power to its high-frequency power. An abnormally high ratio usually indicates an increase in psychological stress load.

[0063] HRV is an abbreviation for Heart Rate Variability. It is an abbreviation for Driver State Index. It is an abbreviation for Vehicle Stability Margin. It is an abbreviation for Environmental Threat.

[0064] In some embodiments of the present invention, the step of classifying scenarios based on comprehensive perception data and identifying the current driving scenario type includes: The comprehensive perception data is input into the neural network model for the first level of risk assessment, outputting a preliminary risk level, including low risk, medium risk, high risk, or passability risk. When the preliminary risk level is high risk, the second level of risk assessment is performed, further subdividing the high risk level according to the dominant risk factors. Among them, if the risk mainly stems from the vehicle's dynamic state approaching its limit, it is judged as a stability risk; if the risk mainly stems from the driver's lack of control ability while the vehicle still has stability, it is judged as a handling risk. Finally, different driving scenario types are output according to different risk results. Among them, the driving scenarios include at least: "normal driving", "on the verge of instability", "instability", and "getting out of trouble".

[0065] like Figure 2 As shown, this process uses comprehensive perception data, namely driver state index, vehicle stability margin, and environmental threat level, as parallel decision factors for synchronous input, and outputs accurate scenario classification through a hierarchical mechanism. First, DSI, VSM, and ET are synchronously input into the neural network model for the first-level risk assessment. The rule base of this model defines the mapping relationship between input combinations and risk levels, and can directly output a preliminary classification of "low risk," "medium risk," "high risk," or "passability risk." Among them, driver state (DSI) serves as the core decision factor, enabling it to dominate risk assessment in the early stages of decision-making. Subsequently, the "high risk" results are further refined by tracing their source. If the risk stems from the vehicle's dynamic state exceeding the tolerance (VSM extremely low), it is judged as an "instability" scenario. If the risk stems from the driver's lack of control ability (DSI extremely high) while the vehicle still has margin, it is judged as a "near instability" scenario driven by "controllability risk." This path highlights the forward-looking role of DSI in risk identification. As for "passability risk," it is directly classified as an "escape from trouble" scenario. For example, in a straight highway scenario, when a vehicle continuously deviates from its lane due to driver microsleep (DSI=0.8), but the vehicle dynamics are normal (VSM=0.9) and the environmental threat is low (ET), a traditional system would misjudge it as "normal driving" because the VSM is normal. However, this invention, through parallel fusion, first identifies "high risk" and traces it back to "maneuverability risk" dominated by DSI, and finally outputs a "near instability" scenario, thereby triggering preventive auxiliary control and obtaining predictive safety capabilities that go beyond vehicle dynamics itself. It can accurately intervene in potential risks caused by human factors and achieve a fundamental leap from "vehicle-side post-event response" to "human-vehicle collaborative pre-event warning".

[0066] The aforementioned neural network model is a pre-trained multilayer perceptron classifier. Its input layer receives normalized DSI, VSM, and ET feature vectors, and its output layer is the classification probability of the comprehensive risk level. This model is trained using a large amount of historical driving data and can learn the complex nonlinear relationship between the aforementioned features and the risk level, achieving efficient forward propagation inference and ultimately outputting the classification result of the comprehensive risk level.

[0067] In some embodiments of the present invention, the step of determining the control requirements under the current driving situation based on the scene classification results includes: The scene classification results are sent to the chassis domain controller. The chassis domain controller calls the preset control strategy corresponding to the driving scene type and executes vehicle control operations based on the received scene classification results. At the same time, the performance evaluation module starts working, monitoring the performance indicators that characterize the current scene classification results and their corresponding system control effects in real time. When the performance indicators meet the preset excellent conditions, the existing scene classification rules and thresholds remain unchanged. When the performance indicators continuously deviate from the expected range, or when the scene classification results switch frequently and are accompanied by high control intervention intensity, the online learning mechanism is triggered. The online learning mechanism collects a data sequence containing comprehensive perception data, scene classification results, control commands, and performance indicators, and generates adjustment amounts for classification rules and thresholds through a lightweight incremental learning algorithm. The adjustment amounts are then used to make minor online updates to the decision parameters of the scene type within the preset safety constraints, so that the updated classification rules and thresholds can be used immediately for subsequent scene judgments, thereby forming a closed-loop adaptive optimization.

[0068] like Figure 3 As shown, specifically, the system sends the scene classification results to the chassis domain controller, which then calls and executes the preset control strategy (such as a weight allocation scheme) that matches the scene. Simultaneously, a parallel performance evaluation module begins working in real time. This module evaluates the effectiveness of the current scene classification and its corresponding control strategy by monitoring a series of key indicators, including but not limited to: the smoothness of the vehicle's dynamic response, the frequency of driver intervention, the effort the system expends to achieve the control objective (such as the cumulative actuator action), and the frequency of scene label switching. If the performance evaluation module detects consistently excellent indicators (e.g., stable vehicle and stable scene labels), it indicates that the current classification rules and control strategy are suitable for the current environment, and the system maintains its current parameter operation.

[0069] Crucially, if the performance evaluation module detects poor control performance or classification oscillations (e.g., under specific operating conditions, the system frequently oscillates between "normal" and "near instability" scenarios, or although classified as "normal," the controller requires frequent and significant intervention to maintain stability), the online learning mechanism is immediately triggered. At this point, the system inputs the most recent serialized driving data (including raw DSI, VSM, and ET data, scenario classification results, control commands, and performance indicators) into a lightweight incremental learning algorithm module (e.g., an optimizer based on online gradient descent or recursive least squares). The core mission of this algorithm module is to analyze historical data and explore the direction of minor adjustments to the current scenario classification rules and thresholds to optimize control performance indicators. The parameter adjustments calculated by the online learning algorithm are passed to the dynamic update module, which is responsible for online, incremental, and safe real-time updates to the membership function thresholds of the fuzzy logic decision-maker's rule base or the decision boundaries of the neural network classifier. This update process follows preset safety constraints to ensure that all adjustments are within physically permissible limits.

[0070] Thus, the updated classification rules and thresholds take effect immediately and are used in subsequent scenario classification decisions, forming a closed-loop benefit system of "scenario classification - control execution - performance evaluation - online learning - rule update - new round of classification". This closed loop enables the entire system to cope with changes in vehicle load, tire wear, differences in driver habits, and unforeseen long-tail scenarios, ultimately achieving a leap from "static preset" to "dynamic adaptation", significantly improving the generalization ability and control precision in complex scenarios.

[0071] In some embodiments of the present invention, after the step of determining the control requirements under the current driving situation based on the scene classification result, the method further includes: Based on the current driving scenario type, the system dynamically configures the priority ranking of multiple control performance indicators according to the safety requirements and control objectives of the scenario, and adjusts the weight allocation in the multi-objective collaborative optimization process based on the priority ranking. The multi-objective collaborative optimization is constructed based on the multi-dimensional deviation between the vehicle's current state and the desired target state. These multi-dimensional deviations include at least path tracking deviation, vehicle stability deviation, and ride comfort-related deviations. The system dynamically adjusts the weight allocation of each type of deviation in the optimization objectives according to the driving scenario, outputting the control command sequence with the minimum overall cost.

[0072] In other words, the system dynamically determines which control objectives are more important in the identified driving scenario (e.g., safety priority or comfort priority), and adjusts the weights of different performance indicators (e.g., path tracking accuracy, vehicle stability, and ride comfort) in the multi-objective optimization accordingly. Finally, by solving an optimization problem with these weighted deviations as the cost function, the system generates a sequence of control commands that minimizes the overall control cost.

[0073] For example, in normal driving scenarios, the control objective is to provide a smooth driving experience with minimal intervention, prioritizing: 1. Driving comfort / economy > 2. Handling > 3. Safety; in near-instability scenarios, the control objective is to actively and smoothly assist the driver in restoring stability and preventing instability, prioritizing: 1. Safety (pre-stability) > 2. Handling > 3. Comfort; in instability scenarios, the control objective will take all necessary measures to quickly and effectively correct the vehicle's posture and stabilize it, prioritizing: 1. Safety (stability) >> 2. Comfort; in getting out of trouble scenarios, the control objective will maximize traction to assist the vehicle in getting out of trouble, prioritizing: 1. Passability > 2. Safety > 3. Comfort.

[0074] Furthermore, the system dynamically configures the priority ranking of multiple control performance indicators based on the safety requirements and control objectives of the scenario, and adjusts the weight allocation in the multi-objective collaborative optimization process based on the priority ranking, including the following steps: The control objective is constructed as an optimization objective function. Its specific form is:

[0075] in, The state error vector includes at least tracking error, stability error, and comfort error. To control the input quantity, it includes all vehicle actuator actions, which at least cover vehicle actuator commands for the braking, steering, drive and suspension systems; This is the state weight matrix; It is the control weight matrix; the solution is to minimize J to achieve the optimal result.

[0076] In one embodiment, based on tracking error e trk Stability error e stab Comfort error e comf The state error vector can be defined as:

[0077] The state weight matrix Q is a diagonal matrix, and the elements on the diagonal are the weight coefficients of the corresponding error terms, as follows:

[0078] The control input vector u is as follows:

[0079] The control weight matrix R is a diagonal matrix, and the elements on the diagonal are the weight coefficients of the corresponding control variables, as follows:

[0080] Where r δ It is the shift weight, r T It is the driving weight, r brake It is the braking weight, r susp It is the suspension weight.

[0081] Of course, the weights mentioned above are not limited to these, and may also include more state weights and control weights.

[0082] In some embodiments of the present invention, the system dynamically assigns priority to various control performance indicators based on the current driving scenario type, and quantifies each control performance indicator into an initial weight vector. It is expressed by the following formula: .

[0083] in, The initial weight represents the i-th control objective, and S represents the driving scenario identifier. The system generates the corresponding initial state weight and control weight based on the current driving scenario identifier through a preset scenario-weight mapping relationship. The weights are used to construct the objective function of the multi-objective optimization problem. The system solves the optimal control command sequence online based on the dynamic weights and sends the first element of the command to each vehicle actuator to achieve multi-system collaborative control.

[0084] In other words, these weighting coefficients are no longer unique fixed values. Instead, after the driving scenario is determined, the system dynamically allocates the priority of various control performance indicators and quantifies them into an initial weight vector. The input to this module comes from the quantized scene identifier S from the upper-layer scene classifier. The process is as follows: the system has a built-in scene-weight mapping function that accepts the scene identifier S and outputs the corresponding initial weight vector. .

[0085] Specifically, the mapping relationship is illustrated below: When the system is identified as experiencing an instability scenario, the stability weight is significantly increased. Reduce the weight of comfort This allows for more aggressive control. When the system is identified as being on the verge of instability, the intensity can be moderately increased. and (Tracking), and maintaining a certain level of [something] This aims to strike a balance between safety and comfort. When the system identifies a normal driving scenario, it increases... They prioritize comfort.

[0086] Based on the dynamic weights determined by the scenario in the above steps, the system uses online rolling optimization based on optimal control theory to minimize the objective function J, thus achieving the optimal control command sequence u (control vector). The dynamic weight coefficients are then adjusted, and the first element is sent to the actuators (braking, steering, drive, and suspension systems, etc.) to achieve coordinated control of the vehicle.

[0087] Furthermore, the detectable feature in this module is that the basic behavioral tendencies of the vehicle's control system (such as steering sensitivity and acceleration response) change measurably and repeatably under different preset driving modes (such as Comfort and Sport). For example, in "Sport" mode, the same steering wheel input produces a greater yaw rate, indicating that steering control is given higher weight.

[0088] In some embodiments of the present invention, the step of dynamically adjusting the weight allocation of each control objective in the decision-making process based on the scenario judgment result and the actuator health status monitoring result includes: Based on the current driving scenario type, the system dynamically assigns priority to multiple control performance indicators and quantifies each control performance indicator into an initial weight vector. When any actuator is detected to have experienced performance degradation or failure, the system proportionally reduces the initial weight of that actuator in the control input vector and renormalizes the weights of the remaining healthy actuators, so that the control tasks originally undertaken by the failed actuator are automatically assigned to the functionally relevant healthy actuators. The multi-objective optimization problem after weight adjustment aims to minimize the overall cost, and generates and issues an updated sequence of control instructions.

[0089] It is understandable that an actuator health status monitoring mechanism is introduced in the dynamic weight coefficient allocation stage. This mechanism ensures that when an actuator fails, the system automatically allocates the control task corresponding to the failed actuator to other functionally related healthy actuators by adjusting and optimizing the weight allocation of the objective function, thereby achieving redundancy and fault tolerance based on control function.

[0090] Furthermore, when any actuator is detected to have experienced performance degradation or failure, the system proportionally decays the initial weight vector corresponding to that actuator in the control input vector and renormalizes the weights of the remaining healthy actuators, so that the control tasks originally undertaken by the failed actuator are automatically allocated to the functionally relevant healthy actuators. This process includes the following steps: Receive initial weight vector and an actuator health status vector H, where each element of the health status vector represents the degree of functionality of the corresponding actuator; determine whether there are non-1 values in the health status vector H; if not, keep the original weights unchanged; if so, proportionally decay the weight components corresponding to each actuator with a health status less than 1, and the decayed weight component is equal to the original weight multiplied by the corresponding health status value;汇总所有衰减后的权重分量形成降级权重向量,并对该降级权重向量进行归一化处理,生成更新后的权重向量;将更新后的权重向量作为后续优化计算的输入,从而调整权重分配。

[0091] As Figure 4 shown, first, input two key input vectors, specifically as follows: (1) The original weight vector for upper-layer scenario classifier decision-making , representing the initial priorities of each control objective.

[0092] (2) The health status vector H of each actuator provided by the actuator health monitoring module. Hi = 0 indicates that the i-th actuator is completely failed; Hi = 1 indicates that its function is normal; 0 < Hi < 1 indicates that its performance is partially degraded.

[0093] Next, make a decision based on the health status vector H of each actuator. The system first checks whether there is an actuator in an unhealthy state (Hi < 1). This is the judgment condition for deciding whether to enter the fault-tolerant adjustment process.

[0094] Then, perform a two-path processing: Path 1 (no fault): If all actuators are healthy, directly adopt the original weights, and the calculated comprehensive cost is the smallest, ensuring the efficient operation of the system in the normal state.

[0095] Path 2 (fault / degradation): Once a fault is detected, start a closed-loop fault-tolerant adjustment process.

[0096] In the fault-tolerant adjustment process, proportionally decay the weight components corresponding to the faulty actuators according to the health status ( ). This specific multiplier structure and the mathematical relationship it defines are different from all traditional redundancy architectures (such as rule-based switching, standby pool switching), and are the necessary and sufficient conditions for realizing the "dynamic self-healing" ability at the control system level, so as to achieve fine-tuning of weight allocation, rather than simple "0-1" switch-like processing.

[0097] Next, normalize the decayed weight vector to ensure that the sum of all weights is still 1, and guide the controller to automatically find a new optimal balance point within the capabilities of the remaining healthy actuators.

[0098] Finally, output the adjusted new weight vector. This is used for the controller optimization solution in the next time step.

[0099] Based on the aforementioned fault-tolerant adjustment process, for example, when the system detects a complete failure or severe performance loss of the braking unit, it will immediately set the weight coefficients of braking-related terms in the objective function of multi-objective collaborative optimization to zero or reduce them to extremely low levels. This is to prevent the issuance of invalid control commands to the failed actuator and avoid system disturbances or energy waste. At the same time, to meet the vehicle's basic deceleration and stopping requirements, the control system activates a cross-actuator redundancy compensation mechanism.

[0100] Specifically, the system can instruct the drive motor to enter reverse or regenerative braking mode, utilizing the electric drive system to provide alternative longitudinal braking force. Furthermore, it can coordinate with the active suspension system to optimize tire-road adhesion utilization by altering the vertical load distribution of the wheels, thereby indirectly increasing the maximum achievable longitudinal braking force. When necessary, the system can also fine-tune the front wheel steering angle or apply active rear wheel steering offset, influencing longitudinal force distribution by changing tire lateral slip characteristics while ensuring lateral stability, further unlocking the potential braking contribution of each chassis subsystem.

[0101] Therefore, the above measures together constitute a multi-degree-of-freedom, cross-functional domain fault-tolerant braking strategy, which enables the vehicle to maintain a certain degree of controllable deceleration capability by integrating and coordinating other healthy actuators even in extreme conditions where the main braking system fails, thus significantly improving driving safety and system robustness.

[0102] In some embodiments of the present invention, the vehicle control method further includes: real-time monitoring of the health status vector of each actuator and transmitting it to the control system via the vehicle bus; when a continuous decay of the health status value of an actuator is detected, the control output of the actuator is proportionally and smoothly reduced according to the change in its health status value; and synchronously calculating the control output of the redundant actuators that need to be added to compensate for the reduced control output of the decaying actuators, so as to ensure that the overall performance indicators of the vehicle remain unchanged or fluctuate within a preset range.

[0103] It is understandable that during fault-tolerant control, the system can identify continuous degradation in actuator performance and generate cooperative response behaviors with identifiable characteristics accordingly. For example, when the health factor of the hydraulic braking subsystem gradually decreases from 1.0 to 0.6 due to thermal fade of the brake disc, the vehicle communication bus can detect that the actual output hydraulic braking force decreases proportionally and smoothly relative to the control command. At this time, the control system does not directly disconnect the subsystem, but dynamically reduces its proportion in the total braking force distribution based on its real-time health status, while coordinating other available redundant actuators (such as regenerative braking of the drive motor, electromechanical braking, etc.) to proportionally increase their output to compensate for the missing braking force.

[0104] This adjustment process manifests as a continuous mapping relationship of "one gain, another loss": a gradual decrease in the output of one actuator corresponds to a gradual increase in the output of another actuator. The two are highly synchronized in the time domain and approximately inversely proportional in amplitude, thus ensuring that key performance indicators such as overall vehicle deceleration and yaw stability remain continuous, without abrupt changes or jitter. This mechanism significantly differs from the hard switching method triggered by fault thresholds in traditional fault-tolerant strategies, avoiding control shocks or system instability caused by sudden actuator start-stop, and achieving a higher level of smooth fault tolerance and performance maintenance.

[0105] This invention incorporates multi-dimensional design features, including driving scenario recognition, control strategy optimization, and cross-system redundancy. In multimodal scene recognition, existing technologies primarily rely on a single data source and static rules for scene segmentation, which can easily lead to misjudgments. These static rules are ill-suited to the complex and ever-changing driving environment and lack collaborative analysis of driver and vehicle status information, failing to accurately identify dynamic human-vehicle-environment needs. Furthermore, existing scene segmentation methods struggle to address diverse requirements under complex operating conditions, resulting in insufficient adaptability and reliability in scene recognition. This invention constructs a dynamic scene classifier through multimodal fusion of driver and vehicle status information, accurately identifying human-vehicle-environment collaborative needs and providing high-confidence scene input for subsequent control.

[0106] In terms of control strategy mapping, most current solutions are based on independent mapping relationships of a single system, directly bound to system parameters, and cannot adapt to real-time changing performance requirements. This invention constructs a scene-actuator dynamic coupling model, which associates the classifier output with indicators such as the health status of the actuator in real time and maps it to a performance indicator optimization function. This can adapt to dynamically changing performance requirements and avoid "over-configuration for under-utilization" or "overload operation".

[0107] In terms of system redundancy design, traditional redundancy designs mainly focus on the signal layer (e.g., dual radars) or the actuator hardware layer (e.g., dual-winding motors). Their redundancy strategies are typically static and isolated, such as directly switching to a backup system when the main system fails. In contrast, this invention proposes a dynamic and flexible functional redundancy method specifically designed for distributed drive platforms. When an actuator fails, the system treats it as a new system constraint and smoothly redistributes the control tasks (e.g., stability control) undertaken by the failed actuator proportionally to other functionally relevant healthy heterogeneous actuators (e.g., drive or braking systems) by dynamically adjusting the weight allocation of the objective function, thereby achieving coordinated control. This method fully utilizes the multi-actuator cooperative characteristics of distributed platforms, improving system fault tolerance and control performance.

[0108] Therefore, this invention establishes a multimodal scene recognition mechanism and develops a dynamic adaptive control mapping algorithm. It introduces the health status information of the execution system (the functional integrity of the execution system) into the weight allocation strategy of multi-objective collaborative optimization, realizing the leap from "information layer redundancy" to "decision layer and functional layer redundancy". This meets the complex and ever-changing driving needs and optimizes the overall performance, significantly improving the system's intelligence level and adaptability to operating conditions. It also provides an innovative solution for the efficient, safe and reliable collaborative control of intelligent chassis control systems.

[0109] It should be noted that "getting out of trouble mode" refers to a specific driving mode that helps the vehicle get out of mud, snow, sand, or other low-traction surfaces by adjusting power distribution, braking logic, and suspension system. Redundancy refers to the additional configuration of duplicate components, functions, or resources in a system to replace and replace existing parts when they fail, ensuring the system's continued stable operation. Stress level is a key indicator of the balance of the autonomic nervous system. An elevated value indicates that sympathetic nervous activity is dominant, the driver is in a state of high stress, psychological stress load is increased, and the risk of driving errors is significantly increased; a decreased value indicates that parasympathetic nervous activity is enhanced, and the driver is in a relaxed state.

[0110] According to a second aspect of the present invention, a vehicle control device includes: a data acquisition unit, an electronic device, an execution unit, and a communication interface. The data acquisition unit is used to acquire driver status data, vehicle dynamics status data, and environmental information data in real time. The electronic device includes a memory and a processor; the memory stores a computer program, and the processor executes a vehicle control method when running the computer program. The execution unit is communicatively connected to the electronic device and is used to receive control commands from the electronic device and execute corresponding vehicle control actions. The execution unit includes at least multiple actuators from a drive system, a braking system, a steering system, and a suspension system. The communication interface is integrated into the electronic device and is used to realize data interaction between the electronic device and the data acquisition unit, the execution unit, and other on-board or off-board devices. The memory includes volatile memory and / or non-volatile memory, and the processor includes at least one of a central processing unit, a digital signal processor, or a dedicated control chip.

[0111] Specifically, the data acquisition unit integrates a variety of high-precision sensors for real-time acquisition of vehicle status and surrounding environment data. These include multimodal sensors such as cameras, lidar, inertial measurement units (IMUs), and physiological sensors, which can capture vehicle motion, surrounding obstacles, road conditions, and driver status information.

[0112] Electronic devices such as Figure 5 As shown, the system mainly includes a memory and a processor. The memory stores computer programs and can include static and dynamic random access memory (SRAM), non-volatile memory, etc. For example, solid-state drives (SSDs) or high-performance hard disks ensure the secure storage of system programs and data. The processor executes the programs stored in the memory and can include a central processing unit (CPU), network processor, digital signal processor, etc. The execution unit is responsible for executing specific vehicle control actions according to the instructions of the computing unit. For example, the drive system adjusts the output of the engine or electric motor to control the vehicle's acceleration and deceleration. The braking system ensures rapid vehicle stabilization in emergency situations through precise braking force distribution. The steering system combines vehicle dynamic data to adjust the steering wheel angle in real time, ensuring the stability of the vehicle's driving direction. The communication interface supports multiple data transmission methods, including wired (such as fiber optic, coaxial cable) and wireless (such as WiFi, Bluetooth, 5G) communication, enabling seamless connection with other devices or systems.

[0113] The hardware device of this invention achieves vehicle stability and safety in complex environments by integrating multimodal sensors, high-performance computing units, and redundant execution systems. The device employs an advanced redundant control architecture to ensure rapid response and effective measures in the event of a fault, providing comprehensive safety assurance for the vehicle.

[0114] In summary, the dynamic weight coefficient allocation method and system based on actuator health status provided by this invention successfully integrates functional safety concepts deep into the core decision-making layer of the control algorithm. Traditional vehicle stability control strategies often rely on fixed weights or preset rules, making it difficult to cope with complex abnormal conditions such as sudden actuator failures, resulting in inherent bottlenecks in system robustness. This invention effectively solves this problem by constructing a closed-loop adaptive adjustment mechanism of "health status perception - dynamic weight decay - control target reconstruction". As shown in the aforementioned flowchart and advantage analysis, the core value of this solution lies in realizing a paradigm shift from static control to dynamic self-healing. The system no longer treats actuator failure as an abnormal event that causes functional interruption, but rather as a new problem that requires online re-optimization. Through the key "link" of weight coefficients, the hardware status of the physical layer and the software strategy of the control layer are organically linked, enabling the system to autonomously and smoothly reconstruct the control logic based on its real-time health status, thereby maintaining the overall performance of the vehicle to the maximum extent even in the event of partial failure. Therefore, this invention not only significantly improves the fault-tolerant control capability and degraded operation performance of intelligent connected vehicles under fault conditions, but also provides technical support for achieving the functional safety goals required for high-level autonomous driving. Its application will effectively enhance the safety and reliability of vehicle operation.

[0115] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0116] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example.

[0117] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A vehicle control method based on multi-source information fusion and actuator health state monitoring, characterized in that, The method comprises the following steps: acquiring driver state data, vehicle dynamics data and environmental information data; performing multi-source information perception and fusion on the driver state data, vehicle dynamics data and environmental information data to generate comprehensive perception data after fusion; classifying the scene according to the comprehensive perception data to identify the current driving scene type; judging the scene based on the scene classification result to determine the control demand under the current driving situation; monitoring the health state of vehicle actuators to evaluate the working performance and reliability of each actuator in real time; dynamically adjusting the weight distribution of each control target in the decision-making process according to the scene judgment result and the actuator health state monitoring result; performing multi-objective collaborative optimization based on the adjusted weight distribution to generate an optimal control strategy; converting the optimal control strategy into control instructions and issuing them to vehicle execution mechanisms to realize the execution of the control instructions.

2. The vehicle control method according to claim 1, characterized by, The driver state data is a driver state index (DRI) is a scalar value between [0, 1] calculated by the following equation:​ wherein, is a fatigue component, is a distraction component, is a stress component; a, b, g are weight coefficients, satisfying + + =1; dynamically dividing the driving risk level based on the numerical interval of the driver state index; The vehicle dynamics data is a vehicle stability margin (S) calculated by the following equation: wherein, is the absolute value of the actual lateral acceleration of the vehicle, is the current road adhesion coefficient, is the current theoretical maximum lateral acceleration of the road, g is the acceleration due to gravity; the numerical interval based on the vehicle stability margin dynamically classifies the vehicle stability state level; The environmental information data is an environmental threat degree (T) , which is calculated by the following formula: wherein, is the minimum curvature radius of the road ahead; is the collision time with the nearest front obstacle; is the current lane width; is the road adhesion coefficient; , , , is a weight coefficient, satisfying dynamically divides the environment threat level based on the numerical interval of the environment threat degree of the environment information data.

3. The vehicle control method according to claim 2, wherein the step of classifying the scene according to the comprehensive perception data to identify the current driving scene type comprises: said fatigue component Based on the eye movement index calculation, the expression is: wherein, is the percentage of time that the eyelid closure exceeds 80% within 60 seconds, is the blink rate (in Hz), 2 is a weighting factor satisfying 1 and and / or the distraction component Based on the line-of-sight deviation and head pose calculation, the expression is: ) wherein, is a time statistic of the line of sight leaving the front single-lane region over a 60-second time period, is a threshold time for the line of sight to leave the front road region; is an angle of the head to deviate from the straight-ahead direction, is a threshold angle for the head to deviate from the straight-ahead direction; and / or said stress component Based on heart rate variability (HRV) calculation, expression is: is the ratio of low frequency power to high frequency power.

4. The vehicle control method according to claim 1, characterized by inputting the comprehensive perception data into a neural network model to perform first-level risk judgment, and outputting a preliminary risk level, including low risk, medium risk, high risk or passability risk; when the preliminary risk level is high risk, performing second-level risk judgment, and subdividing the high risk into risk levels according to risk dominant factors; wherein if the risk mainly originates from the vehicle dynamics state close to the limit, it is determined as stability risk; if the risk mainly originates from the lack of driver control ability while the vehicle still has stability ability, it is determined as control risk; finally, different driving scene types are output according to different risk results; wherein the driving scene types at least include "normal driving", "on the brink of instability", "instability" and "escape from trouble"; and / or the step of judging the scene based on the scene classification result to determine the control demand under the current driving situation comprises: the scene classification result is issued to a chassis domain controller, and the chassis domain controller calls a preset control strategy corresponding to the driving scene type according to the received scene classification result and performs vehicle control operation; at the same time, a performance evaluation module starts to work, and real-time monitoring of performance indicators representing the current scene classification result and the corresponding system control effect is performed; when the performance indicators meet the preset good conditions, the existing scene classification rules and thresholds are maintained unchanged; when the performance indicators continuously deviate from the expected range, or the scene classification result frequently switches with high control intervention intensity, an online learning mechanism is triggered; the online learning mechanism collects data sequences including comprehensive perception data, scene classification results, control instructions and performance indicators, and generates an adjustment amount of classification rules and thresholds through a lightweight incremental learning algorithm; the adjustment amount is used to make slight online updates of the decision parameters of the scene type within a preset safety constraint range, so that the updated classification rules and thresholds are used for subsequent scene judgment in real time, thereby forming a closed-loop adaptive optimization. the step of judging the scene based on the scene classification result to determine the control demand under the current driving situation further comprises:

5. The vehicle control method according to claim 1, characterized by ​ According to the current driving scene type, the system dynamically configures the priority order of multiple control performance indicators according to the safety requirements and control targets of the scene, and adjusts the weight distribution in the multi-objective collaborative optimization process based on the priority order; The multi-objective collaborative optimization is constructed based on the multi-dimensional deviation between the current state of the vehicle and the expected target state, and the multi-dimensional deviation at least includes path tracking deviation, vehicle stability deviation and ride comfort related deviation; the weight distribution of each type of deviation in the optimization target is dynamically adjusted according to the driving scene, and the control instruction sequence with the minimum comprehensive cost is output.

6. The vehicle control method according to claim 5, characterized by The step of dynamically configuring the priority order of multiple control performance indicators according to the safety requirements and control targets of the scene, and adjusting the weight distribution in the multi-objective collaborative optimization process based on the priority order includes: The control target is constructed as an optimization target function (1) in the form of wherein, is a state error vector comprising at least a tracking error, a stability error, a comfort error; is a control input vector comprising all vehicle actuator actions, at least encompassing vehicle actuator commands for braking, steering, drive and suspension systems; is a state weight matrix; is a control weight matrix; solving for Jmin; and / or According to the current driving scene type, the system dynamically allocates the priority order of each control performance index, and quantifies each control performance index into an initial weight vector is represented by the following formula: wherein, an initial weight representing the ith control target, S represents a driving scene identifier, and the system generates a corresponding initial state weight and control weight according to the current driving scene identifier through a preset scene-weight mapping relationship; the weight is used to construct an objective function of a multi-objective optimization problem, the system online solves an optimal control instruction sequence based on the dynamic weight, and the first element of the instruction is issued to each vehicle actuator to realize multi-system collaborative control.

7. The vehicle control method according to claim 1, characterized by The step of dynamically adjusting the weight distribution of each control target in the decision-making process according to the scene judgment result and the actuator health state monitoring result includes: According to the current driving scene type, the system dynamically allocates priority order of multiple control performance indexes, and quantifies each control performance index into an initial weight vector ; When detecting that any actuator has performance degradation or failure, the system proportionally attenuates the initial weight of the actuator corresponding to the control input vector, and re-normalizes the weight of the remaining healthy actuators, so that the control tasks originally undertaken by the failed actuators are automatically distributed to the functionally related healthy actuators; the multi-objective optimization problem after weight adjustment aims to minimize the comprehensive cost, generates an updated control instruction sequence and issues it.

8. The vehicle control method according to claim 7, characterized by, The step of proportionally attenuating the initial weight vector of the actuator corresponding to the control input vector when detecting that any actuator has performance degradation or failure, and re-normalizing the weight of the remaining healthy actuators, so that the control tasks originally undertaken by the failed actuators are automatically distributed to the functionally related healthy actuators includes: receiving an initial weight vector and an actuator health state vector H, wherein elements of the health state vector characterize the degree of functionality of the corresponding actuator; Determine whether there is a non-1 value in the health state vector H; If not, maintain the original weight unchanged; If so, proportionally attenuate the weight component corresponding to each actuator with a health state less than 1, and the attenuated weight component is equal to the original weight multiplied by the corresponding health state value; all attenuated weight components are summarized to form a degraded weight vector, and the degraded weight vector is normalized to generate an updated weight vector; the updated weight vector is used as the input of subsequent optimization calculation, so as to adjust the weight distribution.

9. The vehicle control method according to claim 8, characterized by, The above steps further include: Real-time monitoring of the health state vector of each actuator, and transmission to the control system through the vehicle bus; When detecting that the health state value of an actuator continuously attenuates, the control output of the actuator is proportionally and smoothly reduced according to the change of the health state value; Synchronously calculate the control output amount of the redundant actuators that need to be increased to compensate for the reduced control output of the attenuated actuators, so as to ensure that the overall performance indicators of the vehicle remain unchanged or fluctuate within a preset range.

10. A vehicle control device characterized by comprising: It includes: A data acquisition unit for real-time acquisition of driver state data, vehicle dynamics state data and environmental information data; The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor executes the vehicle control method according to any one of claims 1-9 when running the computer program. An execution unit is communicatively connected to the electronic device, configured to receive a control instruction from the electronic device and execute a corresponding vehicle control action, and the execution unit comprises at least multiple actuators in a drive system, a brake system, a steering system and a suspension system. A communication interface is integrated in the electronic device, configured to realize data interaction between the electronic device, the data acquisition unit, the execution unit and other on-board or off-board devices.

Images