Search

CN-121989943-A - Cooperative anti-skid control method and device for vehicle

CN121989943ACN 121989943 ACN121989943 ACN 121989943ACN-121989943-A

Abstract

The invention discloses a vehicle collaborative anti-skid control method and device, which comprise the steps of obtaining real-time operation data of a vehicle and sending the real-time operation data to a cloud platform, calibrating key dynamic parameters of a cloud digital twin model in real time by utilizing an online machine learning model on the basis of the real-time operation data to obtain real-time calibration parameters, sending the real-time calibration parameters to a vehicle end controller of the vehicle, updating the vehicle end digital twin model by utilizing the real-time calibration parameters at the vehicle end controller, predicting the driving state of the vehicle in a future prediction time domain by utilizing a model prediction control algorithm on the basis of the updated vehicle end digital twin model, solving to obtain an optimal control instruction, and executing the optimal control instruction. The invention changes the traditional passive anti-skid control into active predictive control, can adapt to the dynamic change of key parameters such as road adhesion coefficient and the like in real time, and obviously improves the running safety and stability of the vehicle under extreme road conditions such as ice and snow and the like.

Inventors

  • FAN JIUCHEN
  • SUN XUEMEI
  • ZHANG SHUANG
  • WU DONGDONG

Assignees

  • 北华大学

Dates

Publication Date
20260508
Application Date
20260331

Claims (10)

  1. 1. The cooperative anti-skid control method for the vehicle is characterized by comprising the following steps of: acquiring real-time operation data of a vehicle, wherein the real-time operation data comprises control input data and actual state data of the vehicle; Transmitting the real-time operation data to a cloud platform; On the cloud platform, based on the real-time operation data, real-time calibrating key dynamic parameters of a cloud digital twin model by using an online machine learning model to obtain real-time calibration parameters; Issuing the real-time calibration parameters to a vehicle end controller of the vehicle; Updating a vehicle-end digital twin model by using the real-time calibration parameters at the vehicle-end controller; Based on the updated vehicle-end digital twin model, predicting the running state of the vehicle in a future prediction time domain by adopting a model prediction control algorithm, and solving to obtain an optimal control instruction in the prediction time domain, wherein the optimal control instruction is used for multi-objective collaborative optimization of the running stability of the vehicle; And controlling the actuator of the vehicle according to the optimal control command, wherein the control comprises the cooperative control/joint torque distribution of a chassis execution system of the vehicle.
  2. 2. The method of claim 1, wherein the online machine learning model is a bayesian deep learning network and the real-time calibration parameters are probability distributions of the key dynamic parameters, the probability distributions including a mean and a variance.
  3. 3. The method of claim 2, wherein the Bayesian deep learning network is a Bayesian gated loop unit network, and the key dynamic parameters include at least a tire-road peak attachment coefficient, a vehicle equivalent mass, and a rotational inertia of the vehicle about a Z axis.
  4. 4. The method according to claim 1, wherein the step of calibrating key dynamic parameters of the cloud digital twin model in real time using the online machine learning model specifically comprises: Inputting the control input data to the online machine learning model and the cloud digital twin model simultaneously; Driving the cloud digital twin model to simulate by utilizing parameters output at the moment on the online machine learning model to obtain simulation state data; calculating virtual-real deviation between the actual state data and the simulation state data; Based on the virtual-real deviation, the internal weight of the online machine learning model is updated online through a back propagation algorithm as a loss function, so that the real-time calibration parameters output by the model can minimize the virtual-real deviation.
  5. 5. The method of claim 1, wherein the vehicle-side digital twin model is a seven-degree-of-freedom whole vehicle dynamics model taking nonlinear tire characteristics into account, and wherein the updating the vehicle-side digital twin model is specifically performed by using a tire-road surface peak adhesion coefficient average value in the real-time calibration parameters to update corresponding parameters of a Pacejka tire model in the seven-degree-of-freedom whole vehicle dynamics model in real time.
  6. 6. The method of claim 1, wherein the cost function of the model predictive control algorithm includes at least a comprehensive optimization objective for vehicle yaw stability, tire slip efficiency, and control ride, and wherein the optimal control command is a four-wheel drive torque distribution scheme that minimizes the cost function while satisfying physical constraints of the vehicle.
  7. 7. A cooperative slip control apparatus for a vehicle, comprising: the system comprises a data acquisition module, a control module and a control module, wherein the data acquisition module is configured to acquire real-time operation data of a vehicle, and the real-time operation data comprises control input data and actual state data of the vehicle; the vehicle cloud communication module is configured to send the real-time operation data to a cloud platform and receive real-time calibration parameters issued by the cloud platform; The cloud calibration module is deployed on the cloud platform and is configured to calibrate key dynamic parameters of the cloud digital twin model in real time by utilizing an online machine learning model based on the real-time operation data to obtain the real-time calibration parameters; The model updating module is configured to update a vehicle-end digital twin model in the vehicle-end controller by utilizing the real-time calibration parameters; The prediction control module is configured to predict the running state of the vehicle in a future prediction time domain by adopting a model prediction control algorithm based on the updated vehicle-end digital twin model, and solve to obtain an optimal control instruction in the prediction time domain, wherein the optimal control instruction is used for multi-objective collaborative optimization of the running stability of the vehicle; and the control execution module is configured to control the actuator of the vehicle according to the optimal control instruction, wherein the control comprises cooperative control/joint torque distribution of a chassis execution system of the vehicle.
  8. 8. The apparatus of claim 7, wherein the online machine learning model is a bayesian deep learning network and the real-time calibration parameter is a probability distribution of the key dynamic parameter, the probability distribution comprising a mean and a variance.
  9. 9. The apparatus of claim 7, wherein the cloud calibration module is specifically configured to: Inputting the control input data to the online machine learning model and the cloud digital twin model simultaneously; Driving the cloud digital twin model to simulate by utilizing parameters output at the moment on the online machine learning model to obtain simulation state data; calculating virtual-real deviation between the actual state data and the simulation state data; Based on the virtual-real deviation, the internal weight of the online machine learning model is updated online through a back propagation algorithm as a loss function, so that the real-time calibration parameters output by the model can minimize the virtual-real deviation.
  10. 10. The apparatus of claim 7, wherein the vehicle-side digital twin model is a seven-degree-of-freedom whole vehicle dynamics model accounting for nonlinear tire characteristics, wherein the cost function of the model predictive control module at least comprises a comprehensive optimization objective for vehicle yaw stability, tire slip efficiency and control smoothness, and wherein the optimal control command is a four-wheel drive torque distribution scheme that minimizes the cost function while satisfying physical constraints of the vehicle.

Description

Cooperative anti-skid control method and device for vehicle Technical Field The invention relates to the technical field of joint control of vehicle subunits with different types or different functions, in particular to a cooperative anti-skid control method and device for vehicles. Background With the deep transformation of the automobile industry to electric, intelligent and networking, the active safety performance of vehicles is receiving unprecedented attention. In various active safety technologies, an anti-slip control system is a core function for guaranteeing the running stability of a vehicle under complex and severe road conditions. Currently, anti-skid control systems commonly installed in mass production vehicles mainly include an anti-lock brake system (ABS), a Traction Control System (TCS), and an Electronic Stability Program (ESP) as both integration and sublimation. These systems determine whether there is a risk of excessive slip or lock-up of the wheels by monitoring the rotational speed difference of each wheel, and in turn maintain the longitudinal and lateral stability of the vehicle by adjusting the output torque of the drive motor or applying a braking force to a particular wheel. However, conventional ABS/TCS/ESP systems have their inherent limitations in the technical paradigm. These systems are essentially a passive control based on error feedback, whose logic of operation is a "detect-respond" mode. That is, the control system initiates intervention only after the sensor has captured a significant state deviation (e.g., slip rate exceeding a preset empirical threshold) when significant slip has occurred to the wheel. This post-remediation control system works effectively when the road surface adhesion conditions are good, but its control effect is greatly impaired on ice and snow road surfaces, wet road surfaces or mixed road surfaces (such as ice-water mixtures) where the adhesion conditions are extremely poor and the road surface is suddenly changed. The root cause is that there are physical and algorithmic delays inherent to the overall closed loop process from the detection of anomalies by the sensors, the calculation of decisions by the controller, and the action by the actuators. Under the critical working condition of high-speed running of the vehicle, even a delay of tens of milliseconds can cause the vehicle to miss the optimal control time, so that the posture of the vehicle body enters a unstable state which is difficult to recover, and serious traffic accidents are caused. For example, TCS cannot predict real-time road adhesion limits when controlling the drive wheels, nor cooperate with the steering system to achieve an optimal in-turn trajectory. This non-joint, non-predictive control mode is a fundamental bottleneck limiting the safety performance of existing vehicles under extreme conditions. The prior art has explored some ways to overcome the hysteresis problem described above. For example, researchers have attempted to estimate the peak adhesion coefficient between a tire and a road surface on-line by building a vehicle dynamics model and using state observer techniques such as kalman filtering. The method uses the attachment coefficient as a to-be-estimated state of the system, and uses the error between the actual dynamic response of the vehicle and the model prediction response to carry out recursive estimation. Although this improves the perceived speed of road surface changes to some extent, its accuracy of identification is highly dependent on the accuracy of a pre-set, typically simplified, vehicle model. However, parameters of a real vehicle, such as the mass of the whole vehicle (due to load variations), the centroid position, the tire wear state, etc., are dynamically changing during the life cycle of the vehicle. An off-line calibrated parameter curing model cannot accurately reflect real-time dynamic characteristics of a vehicle, so that deviation and even divergence of adhesion coefficient estimation are caused, and the problem of non-self-adaptability exists. Another class of improvements is that of data driven methods, particularly deep learning techniques. The method comprises the steps of collecting a large amount of vehicle data under different road surfaces in a test field, training a neural network model offline, classifying the current road surface type (such as identifying as 'ice surface' or 'snow surface'), and calling preset control parameters corresponding to the road surface type. The advantage of this approach is that it does not rely on an accurate physical model, but its generalization ability is limited by the completeness of the training dataset. For new road surfaces or complex working conditions which do not appear in the training set, the recognition accuracy of the model can be obviously reduced. More importantly, once the offline training model is deployed to the vehicle end, the capability of the model is solidified, self-e