Vehicle stability self-adaptive control method under low attached road surface multi-source interference

Abstract

The invention discloses a vehicle stability adaptive control method under low attachment road surface multi-source interference, and belongs to the field of vehicle control, and the method comprises the following steps: S1, multi-source interference perception and attachment coefficient prediction; s2, dynamic correction of tire dynamic parameters: correcting tire cornering stiffness and slippage stiffness on line; s3, multi-source interference collaborative observation and estimation: dynamically estimating an interference source by using an enhanced extended state observer; s4, phase plane stability envelope analysis and weight adjustment: dynamically allocating control priorities; and S5, on the premise of considering an interference source, performing dynamic torque distribution and actuator cooperative control: optimizing four-wheel torque by using a quadratic programming algorithm to obtain a torque distribution result. By adopting the vehicle stability adaptive control method under the low-attached road surface multi-source interference, the problems of insufficient modeling precision, weak interference suppression, target conflict and redundancy deficiency in low-attached road surface vehicle stability control are solved, and the trajectory tracking precision and the yaw stability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of vehicle control technology, and in particular to a vehicle stability adaptive control method under multi-source interference on low-adhesion roads. Background Art

[0002] On ice, snow, wet and slippery low adhesion roads (road adhesion coefficient During low-adhesion driving, the friction between the vehicle's tires and the road surface decreases significantly, easily leading to stability issues such as sideslip and tailspin. Furthermore, vehicles in low-adhesion conditions often face the combined effects of multiple interference sources, including crosswinds (>15m / s), sudden changes in road adhesion (such as alternating between icy and dry roads), and brake hydraulic delays (0.8ms / °C). This makes it difficult for traditional control systems to balance path tracking accuracy and yaw stability. For example, when crosswind and sudden changes in adhesion act together, a traditional MPC system can experience a yaw rate overshoot of up to 12° / s, increasing the risk of sideslip by 68%.

[0003] To solve the above problems, the existing technology mainly starts from the following aspects: 1. Tire dynamics modeling: Existing technologies generally use linear tire models such as the magic formula, which simplifies the nonlinear relationship between cornering stiffness and slip rate by assuming a steady state. For example, the magic formula uses a polynomial to fit the relationship between cornering force and slip angle, but ignores the coupling effect between slip rate and slip angle. In addition, traditional models do not consider the temperature variation of tread rubber hardness (-2.3% hardness / °C) and the dynamic characteristics of road adhesion, resulting in poor performance in low-adhesion conditions ( ) The lateral force prediction error exceeds 40%.

[0004] 2. Single-sensor perception and fixed-parameter control strategies: Existing systems rely on single sensors (such as wheel speed and steering wheel angle) such as ABS (anti-lock braking system) and ESP (electronic stability program) for stability control. When disturbances such as crosswind, sudden adhesion changes, and execution delays combine, traditional control strategies struggle to dynamically adjust compensation priorities. For example, when crosswind (>15 m / s) and sudden adhesion changes combine, yaw rate overshoot can reach 12° / s. Traditional PID controllers, with response times exceeding 100 ms, are unable to effectively suppress dynamic deviations. Furthermore, single sensors (such as wheel speed) cannot accurately distinguish between icy and snowy areas (lidar reflection intensity <30) and slippery roads, resulting in insufficient perception accuracy. Some studies have attempted to use fuzzy control to compensate for crosswind disturbances, but the combined suppression rate for multiple sources of disturbances is less than 45%. Furthermore, the temperature-dependent delay characteristics of brake fluid viscosity are not considered, increasing hydraulic response errors by 23%.

[0005] 3. Limitations of actuator redundancy design: Existing redundancy mechanisms are primarily focused on the controller level. For example, steer-by-wire systems utilize dual-winding motors for hardware redundancy, but actuator-level redundancy is lacking. For example, traditional ESP systems lack compensation strategies for single-motor failure or hydraulic failure, resulting in loss of control distances of up to 55 meters in extreme scenarios. While some research has attempted to achieve torque distribution through independent four-wheel drive, the lack of dynamic weighting results in an equal reduction in torque when the dual electric drive axles slip, leading to a 19% increase in yaw moment deviation.

[0006] 4. Insufficient decoupling of control objectives: Existing methods usually use fixed weights to allocate path tracking and stability control objectives, resulting in a decrease in overall performance. For example, in the double lane change test on snow, the fixed weight strategy caused the center of mass slip angle to exceed 5°, triggering a side slip warning. Although existing patents such as CN202411844853A propose a coordinated control strategy for different working conditions, they do not solve the problem of path tracking on low-adhesion roads (lateral deviation <0.2m) and yaw stability ( )’s strong coupling contradiction.

[0007] In summary, existing technologies have significant defects in multi-source interference perception, tire model adaptability, redundant fault tolerance, and control target decoupling. Summary of the Invention

[0008] The purpose of the present invention is to provide a method for adaptively controlling vehicle stability under multi-source interference on low-adhesion roads to solve the above technical problems.

[0009] To achieve the above object, the present invention provides a method for adaptively controlling vehicle stability under multi-source interference on a low-adhesion road surface, comprising the following steps: S1. Multi-source interference perception and adhesion coefficient prediction: The road adhesion coefficient is estimated using collected multimodal environmental data and vehicle dynamics data, and then fused with the road adhesion coefficient predicted using a spatiotemporal convolutional network. S2. Dynamic correction of tire dynamic parameters: Based on the changing trend of the fused road adhesion coefficient prediction value and tire temperature data, the tire cornering stiffness and slip stiffness are corrected online; S3. Collaborative observation and estimation of multi-source interference: Dynamically estimate interference sources using the enhanced extended state observer; S4, Phase plane stability envelope analysis and weight adjustment: Based on the dynamic prediction value of the adhesion coefficient, the Lyapunov stability theory is used to define the phase plane boundary and dynamically allocate control priorities; S5. Dynamic torque distribution and actuator coordinated control: Taking interference sources into consideration, the quadratic programming algorithm is used to optimize the four-wheel torque based on the weighted dynamic distribution results and the corrected tire cornering stiffness to obtain the torque distribution result.

[0010] Preferably, step S1 specifically includes the following steps: S11. Collect multimodal environmental data and vehicle dynamics data, and perform spatiotemporal alignment after filtering noise. The multimodal environmental data includes the point cloud elevation map and road image data in front of the vehicle, as well as the ambient temperature and humidity, and crosswind disturbance torque. The vehicle dynamics data includes the vehicle's roll angle. , pitch angle , vehicle longitudinal speed and vehicle lateral speed ; S12. Compare the reflection intensity value of each point in the point cloud elevation map with a set threshold. If the reflection intensity value is less than the set threshold, the area is determined to be covered by ice and snow and regarded as a low-adhesion road surface. If the reflection intensity value is greater than the set threshold, the area is determined to be a non-low-adhesion road surface. And calculate the road adhesion coefficient of low adhesion road based on point cloud elevation map : ; Where, Indicates the reflection intensity value; At the same time, the road surface image data is input into the YOLOv5 network for road surface texture feature segmentation, and combined with the ResNet-50 classification network to output the road adhesion coefficient of low-adhesion road surface : ; Where, Indicates the probability that low-adhesion road surface is a visual semantic; Collect historical road adhesion coefficient, ambient temperature and humidity, and vehicle dynamics status data, input the collected data into the spatiotemporal convolutional network, and use the spatiotemporal convolutional network to predict the road adhesion coefficient in the next 5 seconds. ; And compensate for road slope disturbance based on vehicle dynamics data: ; Where, Indicates the slope compensation amount; S13, the road adhesion coefficient of the low-adhesion road surface calculated based on the point cloud elevation map , Road adhesion coefficient of low-adhesion road surface based on the output of ResNet-50 classification network And the road adhesion coefficient predicted by the spatiotemporal convolutional network in the next 5 seconds Fusion is performed to obtain the fused road adhesion coefficient : ; Where, 、 、 and are weight coefficients, and .

[0011] Preferably, in step S11, a Velodyne VLS-128 laser radar is used to emit laser pulses to the road surface 50m in front of the vehicle, and the three-dimensional coordinates of the reflected signals are recorded to generate a point cloud elevation map, which includes the three-dimensional coordinates of each point and its reflection intensity value.

[0012] Preferably, step S2 specifically includes the following steps: S21, based on real-time Dynamically correct cornering stiffness to compensate for the slip rate-slip angle coupling effect: ; Where, represents the corrected cornering stiffness; represents the initial cornering stiffness; S22, using a dual Kalman filter to update the slip stiffness attenuation factor and the temperature compensation coefficient in real time based on tire slip rate feedback and tire temperature data; S23. Correct the tire lateral force based on the updated slip stiffness attenuation factor and temperature compensation coefficient and the corrected cornering stiffness.

[0013] Preferably, step S3 specifically includes the following steps: S31. Set the interference sources as crosswind disturbance torque, road adhesion gradient and brake hydraulic delay, and regard the interference sources as expanded states. Construct an augmented state space containing vehicle dynamics state data and expanded states to obtain the interference source estimation value. : ; Where, represents the weight coefficient; represents a low-pass filter, represents the time constant, represents the Laplace variable; Indicates the measured value; represents the output prediction value based on the vehicle dynamics model; Indicates the type of interference source, , corresponding to the crosswind disturbance torque, road adhesion gradient and brake hydraulic delay respectively; S32, considering the spatiotemporal correlation of crosswind, road surface mutation, and actuator error, dynamically updating the weight coefficients of the enhanced extended state observer through unscented Kalman filtering ; S33, output estimated crosswind disturbance torque, road adhesion gradient and brake hydraulic delay.

[0014] Preferably, step S4 specifically includes the following steps: S41. Define the yaw rate boundary using Lyapunov stability theory and the center of mass sideslip angle boundary : ; ; Where, represents the acceleration due to gravity; S42, based on the distance between the vehicle state and the stability boundary The warning levels are divided and the control mode is switched based on the divided warning levels.

[0015] Preferably, in step S42, a dynamic weight adjustment mechanism is set: ; Where, Represents dynamic weight; represents the adjustment factor, and ; represents the critical threshold, and ; when When the , and activate longitudinal acceleration limitation; when It is designated as Level 2 warning. , activate torque distribution and limit the front wheel steering angle change rate to ≤15° / s; when It is designated as Level 3 warning. , switch to stability priority mode and turn off ESP intervention delay.

[0016] Preferably, step S5 specifically includes the following steps: S51. Construct a 19-DOF vehicle model that includes suspension geometric nonlinearity, tire relaxation effects, and motor torque saturation characteristics: ; Where, and Represent the vehicle longitudinal acceleration and vehicle lateral acceleration respectively; Indicates the vehicle mass; 、 、 and Represent the longitudinal forces of the front left, front right, rear left, and rear right wheels respectively; 、 、 and Represents the lateral forces of the front left, front right, rear left, and rear right wheels respectively; represents the yaw angular velocity; represents the derivative of the yaw rate; Indicates that the vehicle is around The moment of inertia of the shaft; 、 、 and Represent the torque of the front left, front right, rear left, and rear right wheels respectively; S52: Minimizing the torque deviation and yaw moment deviation is the goal, and optimizing the torque distribution by considering motor saturation, battery power constraint, and slip ratio constraint. The objective function expression is as follows: ; Where, Indicates wheels The actual torque; and Represents wheels Target torque and maximum torque; Indicates the source of interference, Represent the crosswind disturbance torque and road adhesion gradient respectively and brake hydraulic delay; represents the weight coefficient; Indicates the yaw moment compensation amount; S53. Solve the objective function to obtain the actual torque of the four wheels and yaw moment compensation .

[0017] Preferably, in step S5, redundant fault-tolerant control and actuator compensation are also provided: detecting whether ESP failure or single motor failure occurs, and if so, maintaining stability through differential torque and steering angle compensation based on the torque distribution result.

[0018] Preferably, when a single wheel slip rate > 30% or an ESP failure signal is detected, a counter-yaw torque is generated by the dual motor differential, and the maximum adjustment range is set to ±1200Nm; Based on Actively adjust the front wheel steering angle and limit the steering angle speed to 30° / s.

[0019] Therefore, the present invention adopts the above-mentioned vehicle stability adaptive control method under multi-source interference on low-adhesion road surface, which has the following beneficial effects: 1. High-precision adhesion coefficient estimation: Through multi-source data fusion using lidar (reflection intensity threshold), visual semantic segmentation (YOLOv5+ResNet-50), IMU (inertial unit) slope compensation, and spatiotemporal convolutional network (ST-CNN) prediction, the adhesion coefficient estimation error on low-adhesion roads is controlled to MAE ≤ 0.03, with a prediction time of 5 seconds. Sudden value changes (such as the ice-asphalt interface) solve the misjudgment problem of traditional single sensors (for example, the accuracy of ice and snow recognition at night is improved by 50%); 2. Full-Scene Environment Recognition: Utilizes LiDAR point cloud reflection intensity to accurately identify ice and snow areas. Combined with a vision system, it supplements texture feature recognition under thin ice and strong light conditions. The IMU compensates for slope interference, enabling real-time detection of low-adhesion roads in a temperature range of -40°C to 85°C and a preview distance of 15m. This covers complex working conditions such as ice and snow, slippery conditions, and gravel. 3. Introducing a slip rate-slip angle coupling term, the finite difference method is used to calculate nonlinear tire forces in real time. This breaks through the steady-state assumptions of traditional linear models (such as the magic formula) and improves tire force prediction accuracy in extreme conditions (such as sharp turns on icy surfaces). 4. Enhanced Extended State Observer (EESO) combined with UKF (Unscented Kalman Filter) dynamic weight update to achieve crosswind disturbance torque (range ±200Nm, resolution ±5Nm), road adhesion gradient ( , resolution 0.01 / m), collaborative observation of brake hydraulic delay (compensation accuracy ±2ms), interference estimation accuracy ≥ 92% (RMS error < 0.02), an improvement of 47% over traditional observers; 5. In crosswind (20m / s) + Under sudden changes (0.2 to 0.1), the yaw rate overshoot is reduced from 12° / s to 5.1° / s, reducing the risk of sideslip by 67%. When the hydraulic response delay is greater than 50ms, the control command tracking error is ≤±2%, and the response time is shortened to 4ms, eliminating the loss of control caused by disturbance superposition in traditional systems. 6. Through multimodal perception (lidar + vision + IMU) and spatiotemporal convolutional network (ST-CNN), the system can predict the adhesion coefficient at a preview distance of 15m and support the lateral control requirements of Level 4 autonomous driving. For example, in the snow elk test, the system response speed was 30 times faster than human neural reflexes and the trajectory deviation was reduced by 72%.

[0020] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of the vehicle stability adaptive control method under multi-source interference on low-adhesion road surfaces of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purposes, technical solutions and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, where the same or similar numbers throughout represent the same or similar elements or elements with the same or similar functions.

[0023] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products or devices.

[0024] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] like Figure 1 As shown, a method for adaptively controlling vehicle stability under multi-source disturbances on a low-adhesion road surface comprises the following steps: S1. Multi-source interference perception and adhesion coefficient prediction: The road adhesion coefficient is estimated using collected multimodal environmental data and vehicle dynamics data, and then fused with the road adhesion coefficient predicted using a spatiotemporal convolutional network. Step S1 specifically includes the following steps: S11. Collect multimodal environmental data and vehicle dynamics data, and perform spatiotemporal alignment after filtering noise. The multimodal environmental data includes the point cloud elevation map and road image data in front of the vehicle, as well as the ambient temperature and humidity, and crosswind disturbance torque. The vehicle dynamics data includes the vehicle's roll angle. , pitch angle , vehicle longitudinal speed and vehicle lateral speed ; In step S11, a Velodyne VLS-128 laser radar is used to emit laser pulses to the road surface 50 meters in front of the vehicle, and the three-dimensional coordinates of the reflected signals are recorded to generate a point cloud elevation map. The point cloud elevation map includes the three-dimensional coordinates of each point and its reflection intensity value.

[0026] S12. Compare the reflection intensity value of each point in the point cloud elevation map with a set threshold. If the reflection intensity value is less than the set threshold, the area is determined to be covered by ice and snow and regarded as a low-adhesion road surface. If the reflection intensity value is greater than the set threshold, the area is determined to be a non-low-adhesion road surface. And calculate the road adhesion coefficient of low adhesion road based on point cloud elevation map : ; Where, Indicates the reflection intensity value; At the same time, the road surface image data is input into the YOLOv5 network for road surface texture feature segmentation, and combined with the ResNet-50 classification network to output the road adhesion coefficient of low-adhesion road surface : ; Where, Indicates the probability that low-adhesion road surface is a visual semantic; Collect historical road adhesion coefficient (10 frames, 0.5s interval), ambient temperature and humidity, and vehicle dynamics data, and input the collected data into the spatiotemporal convolutional network to predict the road adhesion coefficient in the next 5s using the spatiotemporal convolutional network. ; And compensate for road slope disturbance based on vehicle dynamics data: ; Where, Indicates the slope compensation amount; S13, the road adhesion coefficient of the low-adhesion road surface calculated based on the point cloud elevation map , Road adhesion coefficient of low-adhesion road surface based on the output of ResNet-50 classification network And the road adhesion coefficient predicted by the spatiotemporal convolutional network in the next 5 seconds Fusion is performed to obtain the fused road adhesion coefficient : ; Where, 、 、 and are weight coefficients, and .

[0027] S2. Dynamic correction of tire dynamic parameters: Based on the changing trend of the fused road adhesion coefficient prediction value and tire temperature data, the tire cornering stiffness and slip stiffness are corrected online; Step S2 specifically includes the following steps: S21, based on real-time Dynamically correct cornering stiffness to compensate for the slip rate-slip angle coupling effect: ; Where, represents the corrected cornering stiffness; represents the initial cornering stiffness; After testing and correction, the lateral force prediction error is The range is reduced to within 12%.

[0028] S22: Using a dual Kalman filter to update the slip stiffness attenuation factor (0.6-1.2) and temperature compensation coefficient (-0.0085 / °C) in real time using tire slip rate feedback and tire temperature data. S23. Correct the tire lateral force based on the updated slip stiffness attenuation factor and temperature compensation coefficient and the corrected cornering stiffness.

[0029] S3. Collaborative observation and estimation of multi-source interference: Dynamically estimate interference sources using the enhanced extended state observer; Step S3 specifically includes the following steps: S31. Set the interference sources as crosswind disturbance torque, road adhesion gradient and brake hydraulic delay, and regard the interference sources as expanded states. Construct an augmented state space containing vehicle dynamics state data and expanded states to obtain the interference source estimation value. : ; Where, represents the weight coefficient; represents a low-pass filter, represents the time constant, represents the Laplace variable; Indicates the measured value; represents the output prediction value based on the vehicle dynamics model; Indicates the type of interference source, , corresponding to the crosswind disturbance torque, road adhesion gradient and brake hydraulic delay respectively; S32, considering the spatiotemporal correlation of crosswind, road surface mutation, and actuator error, dynamically updating the weight coefficients of the enhanced extended state observer through unscented Kalman filtering ; S33, output estimated crosswind disturbance torque, road adhesion gradient and brake hydraulic delay.

[0030] S4, Phase plane stability envelope analysis and weight adjustment: Based on the dynamic prediction value of the adhesion coefficient, the Lyapunov stability theory is used to define the phase plane boundary and dynamically allocate control priorities; Step S4 specifically includes the following steps: S41. Define the yaw rate boundary using Lyapunov stability theory and the center of mass sideslip angle boundary : ; ; Where, represents the acceleration due to gravity; when 、 When , the yaw rate boundary is , the center of mass sideslip angle boundary is .

[0031] S42, based on the distance between the vehicle state and the stability boundary The warning levels are divided and the control mode is switched based on the divided warning levels.

[0032] Preferably, in step S42, a dynamic weight adjustment mechanism is set: ; Where, Represents dynamic weight; represents the adjustment factor, and ; represents the critical threshold, and ; when When the , and activate longitudinal acceleration limitation; when It is designated as Level 2 warning. , activate torque distribution and limit the front wheel steering angle change rate to ≤15° / s; when It is designated as Level 3 warning. , switch to stability priority mode and turn off ESP intervention delay.

[0033] S5. Dynamic torque distribution and actuator coordinated control: Taking interference sources into consideration, the quadratic programming algorithm is used to optimize the four-wheel torque based on the weighted dynamic distribution results and the corrected tire cornering stiffness to obtain the torque distribution result.

[0034] Step S5 specifically includes the following steps: S51, build a system that includes suspension geometric nonlinearity, tire relaxation effect (first-order inertia link, time constant ) and motor torque saturation (peak torque ±1500Nm) and characteristics of the 19-DOF vehicle model: ; Where, and Represent the vehicle longitudinal acceleration and vehicle lateral acceleration respectively; Indicates the vehicle mass; 、 、 and Respectively represent the longitudinal forces of the front left, front right, rear left, and rear right wheels, which are calculated in step S23; 、 、 and Respectively represent the lateral forces of the front left, front right, rear left, and rear right wheels, which are calculated in step S23; represents the yaw angular velocity; represents the derivative of the yaw rate; Indicates that the vehicle is around The moment of inertia of the shaft; 、 、 and Represent the torque of the front left, front right, rear left, and rear right wheels respectively; S52: Minimizing the torque deviation and yaw moment deviation is the goal, and optimizing the torque distribution by considering motor saturation, battery power constraint, and slip ratio constraint. The objective function expression is as follows: ; Where, Indicates wheels The actual torque; and Represents wheels Target torque and maximum torque; Indicates the source of interference, Represent the crosswind disturbance torque and road adhesion gradient respectively and brake hydraulic delay; represents the weight coefficient; Indicates the yaw moment compensation amount; For example, on a split road (left wheel , right wheel ) working condition, the optimizer reduces the torque of the right front wheel by 40% (300→180Nm) and increases the torque of the left rear wheel by 25% (320--400Nm), forming a counter-yaw moment of 12Nm·m.

[0035] S53. Solve the objective function to obtain the actual torque of the four wheels and yaw moment compensation .

[0036] In step S5, redundant fault-tolerant control and actuator compensation are also implemented: ESP failure or single motor failure is detected. If so, stability is maintained through differential torque and steering angle compensation based on the torque distribution results. Specifically, if a single wheel slip rate greater than 30% or an ESP failure signal is detected, a counter-yaw torque is generated through the dual motor differential, with a maximum adjustment range of ±1200Nm. Based on Actively adjust the front wheel steering angle and limit the steering angle speed to 30° / s.

[0037] To prove the present invention, the following working condition verification is further carried out: Double lane shift test on ice and snow: lateral error ≤ 0.25m (53% reduction compared to traditional MPC), yaw rate peak ≤ 6° / s (58% reduction); Braking on ice: braking distance 28m (shortened by 38%), vehicle body deviation angle <3°.

[0038] Crosswind+ Sudden change condition: crosswind 20m / s superimposed When the mutation changes from 0.2 to 0.1, the overshoot of the center of mass slip angle is less than 1.2°, and the sideslip risk is reduced by 67%; Actuator delay compensation: When the hydraulic delay is greater than 50ms, the control command tracking error is ≤±2%, and the response time is shortened to 4ms.

[0039] ESP failure scenario: The ASC / DST module maintains a deceleration of 0.3g, and the loss of control distance is reduced by 42% (from 55m to 32m). Single motor failure: After redistribution of the remaining three motors, the yaw rate deviation is <2° / s and the lateral deviation is <0.5m.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for adaptively controlling vehicle stability under multi-source disturbances on low-adhesion roads, characterized by: The following steps are involved: S1. Multi-source interference perception and adhesion coefficient prediction: The road adhesion coefficient is estimated using collected multimodal environmental data and vehicle dynamics data, and then fused with the road adhesion coefficient predicted using a spatiotemporal convolutional network. S2. Dynamic correction of tire dynamic parameters: Based on the changing trend of the fused road adhesion coefficient prediction value and tire temperature data, the tire cornering stiffness and slip stiffness are corrected online; S3. Collaborative observation and estimation of multi-source interference: Dynamically estimate interference sources using the enhanced extended state observer; S4, Phase plane stability envelope analysis and weight adjustment: Based on the dynamic prediction value of the adhesion coefficient, the Lyapunov stability theory is used to define the phase plane boundary and dynamically allocate control priorities; S5. Dynamic torque distribution and actuator coordinated control: Taking interference sources into consideration, the quadratic programming algorithm is used to optimize the four-wheel torque based on the weighted dynamic distribution results and the corrected tire cornering stiffness to obtain the torque distribution result.

2. The method for adaptively controlling vehicle stability under multi-source interference on low-adhesion roads according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11. Collect multimodal environmental data and vehicle dynamics data, and perform spatiotemporal alignment after filtering noise. The multimodal environmental data includes the point cloud elevation map and road image data in front of the vehicle, as well as the ambient temperature and humidity, and crosswind disturbance torque. The vehicle dynamics data includes the vehicle's roll angle. , pitch angle , vehicle longitudinal speed and vehicle lateral speed ; S12. Compare the reflection intensity value of each point in the point cloud elevation map with a set threshold. If the reflection intensity value is less than the set threshold, the area is determined to be covered by ice and snow and regarded as a low-adhesion road surface. If the reflection intensity value is greater than the set threshold, the area is determined to be a non-low-adhesion road surface. And calculate the road adhesion coefficient of low adhesion road based on point cloud elevation map : ； Where, Indicates the reflection intensity value; At the same time, the road surface image data is input into the YOLOv5 network for road surface texture feature segmentation, and combined with the ResNet-50 classification network to output the road adhesion coefficient of low-adhesion road surface : ； Where, Indicates the probability that low-adhesion road surface is a visual semantic; Collect historical road adhesion coefficient, ambient temperature and humidity, and vehicle dynamics status data, input the collected data into the spatiotemporal convolutional network, and use the spatiotemporal convolutional network to predict the road adhesion coefficient in the next 5 seconds. ; And compensate for road slope disturbance based on vehicle dynamics data: ； Where, Indicates the slope compensation amount; S13, the road adhesion coefficient of the low-adhesion road surface calculated based on the point cloud elevation map , Road adhesion coefficient of low-adhesion road surface based on the output of ResNet-50 classification network And the road adhesion coefficient predicted by the spatiotemporal convolutional network in the next 5 seconds Fusion is performed to obtain the fused road adhesion coefficient : ； Where, 、 、 and are weight coefficients, and .

3. The method for adaptively controlling vehicle stability under multi-source interference on low-adhesion roads according to claim 2, characterized in that: In step S11, a Velodyne VLS-128 laser radar is used to emit laser pulses to the road surface 50 meters in front of the vehicle, and the three-dimensional coordinates of the reflected signals are recorded to generate a point cloud elevation map. The point cloud elevation map includes the three-dimensional coordinates of each point and its reflection intensity value.

4. The method for adaptively controlling vehicle stability under multi-source interference on low-adhesion roads according to claim 2, characterized in that: Step S2 specifically includes the following steps: S21, based on real-time Dynamically correct cornering stiffness to compensate for the slip rate-slip angle coupling effect: ； Where, represents the corrected cornering stiffness; represents the initial cornering stiffness; S22, using a dual Kalman filter to update the slip stiffness attenuation factor and the temperature compensation coefficient in real time based on tire slip rate feedback and tire temperature data; S23. Correct the tire lateral force based on the updated slip stiffness attenuation factor and temperature compensation coefficient and the corrected cornering stiffness.

5. The method for adaptively controlling vehicle stability under multi-source interference on low-adhesion roads according to claim 4, characterized in that: Step S3 specifically includes the following steps: S31. Set the interference sources as crosswind disturbance torque, road adhesion gradient and brake hydraulic delay, and regard the interference sources as expanded states. Construct an augmented state space containing vehicle dynamics state data and expanded states to obtain the interference source estimation value. : ； Where, represents the weight coefficient; represents a low-pass filter, represents the time constant, represents the Laplace variable; Indicates the measured value; represents the output prediction value based on the vehicle dynamics model; Indicates the type of interference source, , corresponding to the crosswind disturbance torque, road adhesion gradient and brake hydraulic delay respectively; S32, considering the spatiotemporal correlation of crosswind, road surface mutation, and actuator error, dynamically updating the weight coefficients of the enhanced extended state observer through unscented Kalman filtering ; S33, output estimated crosswind disturbance torque, road adhesion gradient and brake hydraulic delay.

6. The method for adaptively controlling vehicle stability under multi-source interference on low-adhesion roads according to claim 5, characterized in that: Step S4 The specific steps include: S41. Define the yaw rate boundary using Lyapunov stability theory and the center of mass sideslip angle boundary : ； ； Where, represents the acceleration due to gravity; S42, based on the distance between the vehicle state and the stability boundary The warning levels are divided and the control mode is switched based on the divided warning levels.

7. The method for adaptively controlling vehicle stability under multi-source interference on low-adhesion roads according to claim 6, characterized in that: In step S42, a dynamic weight adjustment mechanism is set: ； Where, Represents dynamic weight; represents the adjustment factor, and ; represents the critical threshold, and ; when When the , and activate longitudinal acceleration limitation; when It is designated as Level 2 warning. , activate torque distribution and limit the front wheel steering angle change rate to ≤15° / s; when It is designated as Level 3 warning. , switch to stability priority mode and turn off ESP intervention delay.

8. The method for adaptively controlling vehicle stability under multi-source interference on low-adhesion roads according to claim 6, characterized in that: Step S5 specifically includes the following steps: S51. Construct a 19-DOF vehicle model that includes suspension geometric nonlinearity, tire relaxation effects, and motor torque saturation characteristics: ； Where, and Represent the vehicle longitudinal acceleration and vehicle lateral acceleration respectively; Indicates the vehicle mass; 、 、 and Represent the longitudinal forces of the front left, front right, rear left, and rear right wheels respectively; 、 、 and Represents the lateral forces of the front left, front right, rear left, and rear right wheels respectively; represents the yaw angular velocity; represents the derivative of the yaw rate; Indicates that the vehicle is around The moment of inertia of the shaft; 、 、 and Represent the torque of the front left, front right, rear left, and rear right wheels respectively; S52: Minimizing the torque deviation and yaw moment deviation is the goal, and optimizing the torque distribution by considering motor saturation, battery power constraint, and slip ratio constraint. The objective function expression is as follows: ； Where, Indicates wheels The actual torque; and Represents wheels Target torque and maximum torque; Indicates the source of interference, Represent the crosswind disturbance torque and road adhesion gradient respectively and brake hydraulic delay; represents the weight coefficient; Indicates the yaw moment compensation amount; S53. Solve the objective function to obtain the actual torque of the four wheels and yaw moment compensation .

9. The method for adaptively controlling vehicle stability under multi-source interference on low-adhesion roads according to claim 8, characterized in that: In step S5, redundant fault-tolerant control and actuator compensation are also provided: whether ESP failure or single motor failure occurs is detected. If so, stability is maintained through differential torque and steering angle compensation based on the torque distribution result.

10. The method for adaptive control of vehicle stability under multi-source interference on low-adhesion roads according to claim 9, characterized in that: When a single wheel slip rate > 30% or an ESP failure signal is detected, a counter-yaw torque is generated through the dual motor differential, and the maximum adjustment range is set to ±1200Nm; Based on Actively adjust the front wheel steering angle and limit the steering angle speed to 30° / s.

Images