Driving anti-skid optimization control method based on vehicle-road cooperation road fusion estimation

Abstract

The invention provides a driving anti-slip optimization control method based on vehicle-road cooperation road fusion estimation, and the method comprises the following steps: S1, fusing vehicle-mounted and roadside visual recognition information, and obtaining a vehicle-road cooperation road state recognition result through employing a weighted voting method in combination with road invariant hypothesis and historical information assistance judgment; s2, adaptively adjusting parameters of a dynamic estimator by using a pavement state recognition result, establishing a sample set based on a pavement fusion estimation result, introducing a fuzzy clustering algorithm of particle swarm optimization to realize self-evolution optimization of the structure and the parameters of the estimator, and outputting a fused attachment coefficient estimation result; and S3, designing a slip rate and wheel acceleration combined control strategy to deal with potential slip in advance, and outputting a control torque result of a slip rate controller. Compared with the prior art, the method integrates visual perception and dynamic information to realize accurate estimation of the pavement state, supports advanced intervention of the slip rate controller before slip, and improves the driving stability of the whole vehicle.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle control, and in particular to a drive anti-skid optimization control method based on vehicle-road cooperative road surface fusion estimation. Background Technology

[0002] The accuracy of road surface estimation is a key factor affecting the effectiveness of vehicle active safety control. If the peak coefficient of adhesion between the tire and the road surface can be accurately obtained, it will not only help improve the response speed and control precision of the traction control system, but also effectively reduce the probability of serious traffic accidents such as skidding, rollover, and rear-end collisions in low-adhesion conditions such as rain, snow, and ice.

[0003] Existing research on adhesion coefficient estimation methods mainly falls into two categories: one is estimation algorithms based on vehicle dynamic response, which typically establish a longitudinal or lateral dynamic model of the vehicle and combine it with observational data such as tire-road slip ratio, sideslip angle, and longitudinal acceleration to deduce the estimated value of the adhesion coefficient. This type of method can utilize existing onboard sensors and model physical meanings clearly. However, the observability of the system decreases when tire excitation is insufficient or when slip ratio and sideslip angle are small. At the same time, under complex operating conditions, the tire model exhibits strong nonlinearity and strong time-varying characteristics, resulting in insufficient stability and real-time performance of the estimation results, making it difficult to apply to the dynamic control requirements of multiple operating conditions.

[0004] The second method is based on direct measurement or visual recognition. This involves acquiring road images or surface feature information using sensors such as cameras, lidar, or microwave radar. It then uses image segmentation, texture analysis, and neural network recognition to infer the road surface type and estimate its adhesion level based on a pre-defined mapping relationship. This type of method has advantages such as fast response, wide estimation range, and no reliance on tire excitation. However, it is easily affected by environmental changes such as sunlight, precipitation, and dirt occlusion, and typically only obtains a statistically significant adhesion level, making it difficult to characterize the actual local physical state of the vehicle's contact with the ground.

[0005] In addition, most existing active safety control systems (such as ABS and TCS) adopt a reactive control strategy, that is, intervention control is only triggered after the vehicle skids. There is an unavoidable time delay, which makes it impossible to predictively intervene in potential skid states, thus limiting the system's ability to utilize adhesion under extreme conditions.

[0006] Therefore, there is an urgent need for a new estimation and control method that can integrate visual perception and vehicle dynamics information, while maintaining estimation accuracy and real-time performance, to achieve robust identification of adhesion status and serve the early intervention of anti-skid control systems. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a driving anti-skid control optimization method based on vehicle-road cooperative road surface fusion estimation.

[0008] Technical solution A method for optimizing drive anti-skid control based on vehicle-road cooperative road surface fusion estimation includes the following steps: S1. By integrating vehicle-mounted and roadside visual recognition information, and using a weighted voting method combined with the assumption of road surface invariance and historical information to assist in the judgment, the road surface condition recognition result of vehicle-road cooperation is obtained. S2. Adaptively adjust the dynamic estimator parameters using the road surface condition recognition results output in step S1. At the same time, establish a sample set based on the road surface fusion estimation results, introduce a particle swarm optimization fuzzy clustering algorithm to realize the self-evolution optimization of the estimator structure and parameters, and output the fused peak adhesion coefficient estimation results. S3. Using the fused estimated road surface adhesion information as a feedforward input, a prediction and judgment mechanism is introduced to design a joint control strategy of slip ratio and wheel acceleration to respond to potential slip in advance, and the control torque result of the slip ratio controller is output.

[0009] Specifically, step S1 is as follows: S11. The vehicle-mounted camera and roadside camera acquire images of the target road area and input them into the trained road surface classification model to obtain the road surface recognition results from the vehicle and roadside. S12. Map the road surface recognition results of the vehicle end and the road end to spatial coordinates, corresponding to the grid area under a unified ground fixed coordinate system. At the same time, use a weighted voting mechanism to assign weights to each recognition source, and output the predicted value of the road surface state at the current grid position according to the category probability. S13. Introduce the assumption of road surface invariance to correct the spatial continuity of grid recognition results that change abruptly in a short period of time; combine historical recognition information and update the recognition probability of recent frames based on model reconstruction to achieve temporal consistency filtering and stable output of recognition results.

[0010] S14. Output the final vehicle-road cooperative road surface state recognition result.

[0011] Furthermore, the road surface classification model in step S11 is constructed through the following process: Construct a road surface classification dataset containing images of different road conditions such as dry, wet, and icy / snowy conditions; The UNet network structure is used to segment road surface regions in images. The ShuffleNet network was trained end-to-end based on the road surface classification dataset to complete road surface type identification. Establish a mapping model between road surface type and its corresponding adhesion coefficient.

[0012] Furthermore, the specific process of fusing the road surface recognition results in step S12 includes the following steps: The positions of the vehicle-end and road-end recognition results are transformed by Universal Transverse Mercator (UTM) projection and mapped to a grid area in a fixed ground coordinate system with a set resolution. Trust weights are assigned to the recognition results within each grid cell, with different weights given based on factors such as reliability, viewing resolution, and historical recognition accuracy. The probability outputs of road surface conditions from various sources are weighted and summed, and a weighted voting mechanism is used to output the current road surface condition prediction result for each grid cell.

[0013] Furthermore, the specific process of fusing the road surface recognition results in step S13 includes the following steps: Extract the raster-level multidimensional probability vector output from step S12, and normalize it using the following Sigmoid function to obtain the confidence probability value for each type of road surface condition: Road surface condition is a slowly changing process. By introducing the assumption that the road surface is invariant, spatial continuity correction is performed on the grid recognition results that change abruptly in a short period of time. The model is reconstructed using the following formula, historical frame information is introduced in the time dimension for sliding update and consistency verification, and the grid recognition results are mapped to the corresponding virtual road blocks in the spatial dimension. The confidence vector is reconstructed in reverse to enhance the expression of subdivided features and complete the data-level fusion.

[0014] In the formula, for Road surface types at all times The measurement update probability; for Road surface types at all times The predicted update probability; for Road surface types at all times The predicted update probability; for The road surface type of the i-th road block in the time-mapping image is The estimated probability; This represents the actual total number of road blocks. The expected total number of road blocks; It is a relaxation factor; for Time-based road surface estimation and prediction update values; for Time-based road surface estimation and prediction update values; , , This is the probability threshold that triggers a change in the estimated value.

[0015] Furthermore, the dynamics estimator and fusion estimator in step S2 are designed as follows: The tire longitudinal force estimator and the longitudinal road surface estimator are designed as follows.

[0016] In the formula, For tire longitudinal force estimation, To obtain the adhesion coefficient by substituting the road surface estimate and slip ratio into the tire model fitting formula, For the gain of the longitudinal force estimator, This is the numerical solution for the peak tire-road adhesion coefficient obtained by solving the tire model fitting formula based on the longitudinal force estimate and slip ratio. This is an estimate of the peak tire-road adhesion coefficient (longitudinal excitation). For the road surface estimator gain, These are intermediate variables used in the process of constructing the estimator.

[0017] Therefore, based on the estimation of tire lateral force and self-aligning torque, a lateral road surface estimation algorithm is designed.

[0018] In the formula, For steering wheel angle, To improve the torque-to-rotation ratio of the motor to the kingpin, This is the rotation ratio of the steering wheel torque to the steering torque at the kingpin. Steering wheel torque, To assist the motor torque, A and B These are the fitting parameters for a second-order system. The main return torque. The estimated value of the normalizing moment is calculated based on the vertical load and the sideslip angle. For wheel mechanical trail, This is an estimate of the lateral acceleration. and For the estimator gain, This is an estimate of the peak tire-road adhesion coefficient (lateral force excitation).

[0019] Based on the friction ellipse, the following assumptions can be made regarding the longitudinal and lateral adhesion limits of the tire.

[0020] In the formula, For longitudinal adhesion limit, For lateral adhesion limit, Wheel slip angle The lateral adhesion limit of the contribution, Camber angle The lateral adhesion limit of the contribution, The peak adhesion coefficient, and The value is related to the vehicle's status.

[0021] Based on the above analysis, a dynamic fusion pavement estimator is designed as shown in the following equation, and the formula in the original estimator is modified.

[0022] In the formula, , for and The estimated value.

[0023] The vision and dynamics fusion estimator is designed as follows: In the formula, This is a visual estimate of the road surface adhesion coefficient. This is the road surface estimation result from the previous moment. It is a dynamic estimation of gain that satisfies the local Lipchitz condition. It is a gain adjustment coefficient greater than 0. δ It is the boundary layer thickness. It is the road surface estimate after incorporating visual information.

[0024] Furthermore, step S2 specifically includes: S21. Calculate the boundary layer thickness (i.e., the initial value of the boundary layer thickness) using the road surface state information identified in step S1, and build a dynamic estimator based on this to output the road surface estimate after fusing visual information.

[0025] S22. Construct a fusion estimation model based on fuzzy C-means clustering (FCM), using the latest historical estimation results to form a sample set X, and incorporating labeling information by combining a time decay factor. Define the objective function as follows: In the formula, c is the number of cluster centers, n is the total number of data points, and u ij For membership degree, As a weighted index, d ij Let be the Euclidean distance between the j-th data point and the cluster center, and t be the distance of the data point from the current time label. It is a Lagrange multiplier and also a time decay factor. For the value of the j-th data point, Let be the value of the i-th cluster center.

[0026] The cluster centers and membership degrees are updated as follows: S23. Based on the boundary layer thickness information obtained in step S21, and combined with the classification results of the cluster centers in step S22, the boundary layer thickness is further adaptively updated. Specifically, the membership degree is greater than a threshold. Data points sample points form a set The length is denoted as Boundary layer thickness It can be calculated using the following formula: in It is a cost function; and This represents the number of data points, with values ​​greater than and less than the cluster centers, respectively. It is the proportionality coefficient.

[0027] Furthermore, step S3 specifically includes: S31. Using visual road surface recognition information as feedforward input, calculate the reference slip ratio and the actual slip ratio, while also incorporating wheel acceleration, thus expressing the controlled variable as the following formula: when At that time, the controlled variable is the wheel slip ratio. ,when At that time, the controlled variable is the equivalent wheel acceleration. .when At this time, the controlled variable is a linear combination of the two. Substituting the wheel slip ratio and the reference slip ratio into the above equation, the output is... and .

[0028] S32. Using a conditional integral controller with feedforward, calculate the controller output torque as follows: in , It is an integral term. for initial value, The coefficient before the integral term, The thickness of the synovial surface. This is the upper limit of the driving torque. The actual required torque after slip ratio control intervention; S33. Design the following control strategy based on the visual fusion results between the vehicle-mounted camera and the roadside camera. Predictive slip ratio controller output torque ,when At this time, the system performs slip ratio control, and the motor outputs a predicted control torque. Otherwise, then let Simultaneously, the following logic is executed: calculate the driver's required torque. Output torque of slip ratio controller If the ratio exceeds the set upper limit threshold... The system performs slip ratio control, and the motor outputs control torque; if this criterion is not triggered, the slip ratio controller enters a preload state, calculates and saves the slip ratio controller output torque. When the traditional slip criterion is met, the system immediately switches to preload control mode and quickly outputs the slip ratio controller output torque. This enables a low-latency response to the slip state.

[0029] Traditional slip criteria are based on wheel angular acceleration With longitudinal acceleration The difference, and the preload torque Torque required by current drivers The ratios are used to construct angular acceleration and torque criteria, respectively. For the angular acceleration flag, the flag is set in the initial state. ,when And duration At that time, the flag position At this point, the system will continuously monitor whether the conditions are met. And duration If satisfied, then the flag bit If the above-mentioned conditions are not triggered, the flag state remains unchanged. For the torque flag, the flag is set in the initial state. ,when At that time, the flag position At this point, the system will continuously monitor whether the conditions are met. If satisfied, then the flag bit In all other cases, the flag state remains unchanged. The final torque output is determined by the combined state of the two flags, as shown in the following logic: Beneficial effects This invention integrates visual perception and dynamic information to achieve accurate estimation of road surface conditions. At the same time, it relies on the road surface condition information obtained in advance by visual perception to help judge the risk of future slippage and supports the early intervention of the slip ratio controller before slippage occurs, thereby improving the driving stability of the whole vehicle. Attached Figure Description

[0030] Figure 1 This is a flowchart of the drive anti-skid control optimization method based on vehicle-road cooperative road surface fusion estimation of the present invention; Figure 2 This is a schematic diagram of a grid area in a fixed ground coordinate system according to an embodiment of the present invention; Figure 3 This is a flowchart of the intervention judgment mechanism of the drive slip ratio controller in this invention. Detailed Implementation

[0031] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0032] Example like Figure 1 As shown, this invention provides a driving anti-skid control optimization method based on vehicle-road cooperative road surface fusion estimation, characterized by the following steps: Step S1: Integrate vehicle-mounted and roadside visual recognition information, and use a weighted voting method combined with the road surface invariance assumption and historical information to assist in the judgment, thereby obtaining the road surface condition recognition result of vehicle-road cooperation. Specifically: Construct a road surface classification dataset containing images of different road conditions, including dry, wet, and icy / snowy conditions; The UNet network structure is used to segment the road surface region in the image and extract road surface image patches as image preprocessing for road surface type recognition. The ShuffleNet network was trained end-to-end based on a classification dataset to complete road surface type recognition. A mapping model between road surface type and its corresponding adhesion coefficient was established, and the road adhesion coefficient values ​​for dry road surface, wet road surface and icy / snowy road surface were set to 0.9, 0.7 and 0.2 respectively; Vehicle-mounted cameras and roadside cameras acquire images of the target road area and input them into a trained road surface classification model to obtain road surface recognition results from both the vehicle and roadside. The location of the road surface recognition result is subjected to Universal Transverse Mercator (UTM) projection transformation and mapped to a grid area in a fixed ground coordinate system of a set resolution. Let the boundary of this area in the fixed ground coordinate system be... The grid size is set to 0.5 m × 0.5 m, resulting in a resolution of [resolution value missing]. Each grid cell needs to contain the range of peak adhesion coefficients at that location. and the mathematical expectation estimate of the peak adhesion coefficient (Point estimate), such as Figure 2 As shown; A trust weight is assigned to the recognition result in each grid cell, and different weights are assigned according to factors such as reliability, viewing resolution, and historical recognition accuracy. The probability outputs of each source road surface state (in this embodiment, the road surface recognition results from the vehicle end and the road end) are weighted and summed according to the trust weight, and the current road surface state prediction result of each grid cell is output using a weighted voting mechanism.

[0033] The road surface condition prediction results are then extracted and normalized using the following Sigmoid function to obtain the confidence probability value for each type of road surface condition: The model is reconstructed using the following formula: historical frame information is introduced in the time dimension for sliding updates and consistency checks; in the spatial dimension, the raster recognition results are mapped to the corresponding virtual road blocks; and confidence vectors are reconstructed in reverse to enhance the expression of subdivided features, thus completing data-level fusion. The final visual recognition result is output.

[0034] In the formula, for Road surface types at all times The measurement update probability; for Road surface types at all times The predicted update probability; for Road surface types at all times The predicted update probability; for The road surface type of the i-th road block in the time-mapping image is ( The estimated probability of ). This represents the actual total number of road blocks. The expected total number of road blocks; It is a relaxation factor; for Time-based road surface estimation and prediction update values; for Time-based road surface estimation and prediction update values; , , This is the probability threshold that triggers a change in the estimated value.

[0035] In this embodiment, = 10, = 0.3, = = = 0.7.

[0036] Step S2: Adaptively adjust the dynamic estimator parameters using the road surface condition recognition results output in Step S1. At the same time, establish a sample set based on the road surface fusion estimation results, introduce a particle swarm optimization fuzzy clustering algorithm to realize the self-evolution optimization of the estimator structure and parameters, and output the fused peak adhesion coefficient estimation results.

[0037] The dynamics estimator includes: a longitudinal dynamics estimator, a lateral dynamics estimator, and a fusion estimator.

[0038] Specifically: Longitudinal dynamics estimator: The dynamics estimator and fusion estimator are built based on the following formulas, and the road surface estimate is output after fusing visual information.

[0039] The slip ratio is obtained by the following formula.

[0040] In the formula, The angular velocity of the wheel. The radius of the wheel's rolling motion. The longitudinal velocity at the center of the wheel, λ This refers to the wheel slip ratio.

[0041] To address the issues of insufficient fitting accuracy and unclear physical meaning of parameters in the Burckhardt model, the number of parameters in the original model was increased from 3 to 5. This improved the model fitting accuracy without significantly increasing the model complexity. The improved Burckhardt tire model is shown below and will serve as the fitting formula for subsequent tire models.

[0042] In the formula, In order to utilize the adhesion coefficient, The peak adhesion coefficient, The fitting parameters affect the shape of the curve. It is a symbolic function.

[0043] The tire longitudinal force and peak adhesion coefficient interconnected perturbation estimator designed based on the tire model and single-wheel dynamics model is shown below.

[0044] In the formula, For tire longitudinal force estimation, To obtain the adhesion coefficient by substituting the road surface estimate and slip ratio into the tire model fitting formula, For the gain of the longitudinal force estimator, This is the numerical solution for the peak tire-road adhesion coefficient obtained by solving the tire model fitting formula based on the longitudinal force estimate and slip ratio. This is an estimate of the peak tire-road adhesion coefficient (longitudinal excitation). For the road surface estimator gain, These are intermediate variables used in the process of constructing the estimator.

[0045] Lateral dynamics estimator: Based on the estimation of tire lateral force and self-aligning moment, a lateral road surface estimation algorithm is designed.

[0046] In the formula, For steering wheel angle, The angular velocity of the steering wheel. The steering wheel angular acceleration, To improve the torque-to-rotation ratio of the motor to the kingpin, This is the rotation ratio of the steering wheel torque to the steering torque at the kingpin. Steering wheel torque, To assist the motor torque, A and B These are the fitting parameters for a second-order system. The main return torque. The estimated value of the normalizing moment is calculated based on the vertical load and the sideslip angle. For wheel mechanical trail, This is an estimate of the lateral acceleration. and For the estimator gain, This is an estimate of the peak tire-road adhesion coefficient (lateral force excitation).

[0047] Based on the friction ellipse, the following assumptions about the longitudinal / lateral adhesion limits of the tire can be established.

[0048] In the formula, For longitudinal adhesion limit, For lateral adhesion limit, Wheel slip angle The lateral adhesion limit of the contribution, Camber angle The lateral adhesion limit of the contribution, The peak adhesion coefficient, and The value is related to the vehicle's status.

[0049] Fusion Estimator: Based on the above analysis, a dynamic fusion pavement estimator is designed as shown in the following formula, and the formula in the original estimator is modified.

[0050] In the formula, , for and The estimated value.

[0051] The vision and dynamics fusion estimator is designed as follows: In the formula, This is a visual estimate of the road surface adhesion coefficient. This is the road surface estimation result from the previous moment. It is a dynamic estimation of gain that satisfies the local Lipchitz condition. It is a gain adjustment coefficient greater than 0. The peak adhesion coefficient range is regarded as the boundary layer, and the distance from the point estimate to the boundary layer is the boundary layer thickness, denoted as δ. It is the road surface estimate after incorporating visual information.

[0052] In this embodiment, The value is set to 2.0, the adhesion coefficient range for dry road surfaces is set to [0.80, 1.00], and the adhesion coefficient range for wet road surfaces is set to [0.60, 0.80], with δ set to 0.1.

[0053] A sample set X is established based on the road surface fusion estimation results.

[0054] A fusion estimation model based on parameter self-calibrating fuzzy C-means clustering (FCM) optimized by particle swarm optimization (PSO) is constructed. The latest historical estimation results are used to form a sample set X, and labeling information is introduced by combining a time decay factor. The objective function is defined as follows: In the formula, c is the number of cluster centers, n is the total number of data points, and u ij For membership degree, As a weighted index, d is set to 2.0 in this embodiment. ij Let be the Euclidean distance between the j-th data point and the cluster center i, and t be the distance of the data point from the current time label. Since the Lagrange multiplier is also the time decay factor, in this embodiment it is set to 0.01. For the value of the j-th data point, Let be the value of the i-th cluster center.

[0055] The cluster centers and membership degrees are updated as follows: Finally, based on the set boundary layer thickness value δ and the classification results of the cluster centers, the boundary layer thickness is further adaptively updated. Specifically, the membership degree is greater than a threshold. Data points sample points form a set The length is denoted as Boundary layer thickness It can be calculated using the following formula: in It is a cost function; and This represents the number of data points, with values ​​greater than and less than the cluster centers, respectively. It is a proportionality coefficient, and in this embodiment, it is set to 0.85.

[0056] Step S3: Using the fused estimated road adhesion information as feedforward input, a prediction and judgment mechanism is introduced to design a joint control strategy of slip ratio and wheel acceleration to proactively address potential slippage, and the control torque result of the slip ratio controller is output. Specifically: Using the estimated road adhesion coefficient from step S2 as the feedforward input, the reference slip ratio and the actual slip ratio are calculated, while also incorporating wheel acceleration. The controlled variable is then expressed as follows: when At that time, the controlled variable is the wheel slip ratio. ,when At that time, the controlled variable is the equivalent wheel acceleration. .when At this time, the controlled variable is a linear combination of the two. Substituting the wheel slip ratio and the reference slip ratio respectively, the output is... and .

[0057] Construct a conditional integral controller with feedforward using the following formula, and calculate the controller output torque. .

[0058] in This refers to the wheel speed tracking error (i.e., the error between the reference wheel speed and the actual wheel speed). It is the upper bound of uncertainty. for initial value, The coefficient before the integral term, The thickness of the synovial surface. This is the upper limit of the driving torque. The actual required torque after slip ratio control intervention.

[0059] In this embodiment The value is 10. Takes the value 1. The value is 3200.

[0060] Design the following control strategy: Based on the visual fusion results of vehicle-mounted cameras and roadside cameras The reference slip ratio value under the corresponding state is obtained, and the output torque of the slip ratio controller is predicted using a conditional integral controller with feedforward. ,when At this time, the system performs slip ratio control, and the motor outputs a predicted control torque. ; Otherwise, set the preload torque. Simultaneously execute such Figure 3 The judgment strategy shown: Calculate the required torque for the driver Output torque of slip ratio controller If the ratio exceeds the set upper limit threshold... The system performs slip ratio control, and the motor outputs control torque. ; Otherwise, the slip ratio controller enters a preload state, calculates and saves the slip ratio controller output torque. When the traditional slip criterion is met, the system immediately switches to preload control mode and quickly outputs the slip ratio controller output torque. This enables a low-latency response to the slip state.

[0061] The traditional slip criterion is based on wheel angular acceleration. With longitudinal acceleration The difference, and the preload torque Torque required by current drivers The ratios of these ratios are used to construct angular acceleration criteria. ) and torque criterion ( ).

[0062] The angular acceleration flag is set in the initial state. ,when And duration At that time, the flag position At this point, the system will continuously monitor whether the conditions are met. And duration If satisfied, then the flag bit If the above judgment conditions are not triggered, the flag state remains unchanged.

[0063] The torque flag is set in the initial state. ,when At that time, the flag position At this point, the system will continuously monitor whether the conditions are met. If satisfied, then the flag bit In all other cases, the flag state remains unchanged. The final torque output is determined by the combined state of the two flags, as shown in the following logic: This embodiment , The value is 1.05. The value is 0.95, and S is 1. , Value is 30ms.

[0064] The above description is merely a description of preferred embodiments of this application and is not intended to limit the scope of this application in any way. Any changes or modifications made by those skilled in the art based on the above-disclosed technical content should be considered as equivalent and valid embodiments and fall within the scope of protection of the technical solution of this application.

Claims

1. A method for optimizing drive anti-skid control based on vehicle-road cooperative road surface fusion estimation, characterized in that, Includes the following steps: S1. By integrating vehicle-mounted and roadside visual recognition information, and using a weighted voting method combined with the assumption of road surface invariance and historical information to assist in the judgment, the road surface condition recognition result of vehicle-road cooperation is obtained. S2. Adaptively adjust the dynamic estimator parameters using the road surface condition recognition results output in step S1. At the same time, establish a sample set based on the road surface fusion estimation results, introduce a particle swarm optimization fuzzy clustering algorithm to realize the self-evolution optimization of the estimator structure and parameters, and output the fused peak adhesion coefficient estimation results. S3. Using the fused estimated road surface adhesion information as a feedforward input, a prediction and judgment mechanism is introduced to design a joint control strategy of slip ratio and wheel acceleration to respond to potential slip in advance, and the control torque result of the slip ratio controller is output.

2. The method for optimizing drive anti-skid control based on vehicle-road cooperative road surface fusion estimation according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11. The vehicle-mounted camera and roadside camera acquire images of the target road area and input them into the trained road surface classification model to obtain the road surface recognition results from the vehicle and roadside. S12. Map the road surface recognition results of the vehicle end and the road end to spatial coordinates, corresponding to the grid area under a unified ground fixed coordinate system. At the same time, use a weighted voting mechanism to assign weights to each recognition source, and output the predicted value of the road surface state at the current grid position according to the category probability. S13. Introduce the assumption of road surface invariance to correct the spatial continuity of grid recognition results that change abruptly in a short period of time; combine historical recognition information and update the recognition probability of recent frames based on model reconstruction to achieve temporal consistency filtering and stable output of recognition results. S14. Output the final vehicle-road cooperative road surface state recognition result.

3. The method for optimizing drive anti-skid control based on vehicle-road cooperative road surface fusion estimation according to claim 2, characterized in that, In step S11, the road surface classification model is constructed through the following process: Construct a road surface classification dataset containing images of different road conditions such as dry, wet, and icy / snowy conditions; The UNet network structure is used to segment road surface regions in images. The ShuffleNet network was trained end-to-end based on the road surface classification dataset to complete road surface type identification. Establish a mapping model between road surface type and its corresponding adhesion coefficient.

4. The method for optimizing drive anti-skid control based on vehicle-road cooperative road surface fusion estimation according to claim 2, characterized in that, In step S12, the specific process of fusing road surface recognition results includes the following steps: The positions of the vehicle-end and road-end recognition results are transformed by a general transverse Mercator projection and mapped to a grid area in a fixed ground coordinate system with a set resolution. A trust weight is assigned to the recognition result in each grid cell, and different weights are assigned according to factors such as reliability, viewing resolution, and historical recognition accuracy. The probability outputs of road surface conditions from various sources are weighted and summed, and a weighted voting mechanism is used to output the current road surface condition prediction result for each grid cell.

5. The method for optimizing drive anti-skid control based on vehicle-road cooperative road surface fusion estimation according to claim 2, characterized in that, In step S13, the historical information correction mechanism includes the following steps: Extract the raster-level multidimensional probability vector output from step S12, and normalize it using the following Sigmoid function to obtain the confidence probability value for each type of road surface condition: Road surface condition is a slowly changing process. By introducing the assumption that the road surface is invariant, spatial continuity correction is performed on the grid recognition results that change abruptly in a short period of time. The model is reconstructed using the following formula, historical frame information is introduced in the time dimension for sliding update and consistency verification, and the grid recognition results are mapped to the corresponding virtual road blocks in the spatial dimension. The confidence vector is reconstructed in reverse to enhance the expression of subdivided features and complete the data-level fusion. in, for Road surface types at all times The measurement update probability; for Road surface types at all times The predicted update probability; for Road surface types at all times The predicted update probability; for The road surface type of the i-th road block in the time-mapping image is The estimated probability; This represents the actual total number of road blocks. The expected total number of road blocks; It is a relaxation factor; for Time-based road surface estimation and prediction update values; for Time-based road surface estimation and prediction update values; , , This is the probability threshold that triggers a change in the estimated value.

6. The method for optimizing drive anti-skid control based on vehicle-road cooperative road surface fusion estimation according to claim 1, characterized in that, The specific steps of step S2 are as follows: S21. Calculate the boundary layer thickness using the road surface state information identified in step S1, and build a dynamic estimator based on this to output the road surface estimate after fusing visual information. S22. Construct a fusion estimation model based on fuzzy C-means clustering, using the latest historical estimation results to form a sample set X, and incorporating labeling information by combining a time decay factor. Define the objective function as follows: In the formula, c is the number of cluster centers, n is the total number of data points, and u ij For membership degree, As a weighted index, d ij Let be the Euclidean distance between the j-th data point and the cluster center, and t be the distance of the data point from the current time label. It is a Lagrange multiplier and also a time decay factor. For the value of the j-th data point, The value of the i-th cluster center; Cluster centers and membership degrees are updated as follows: S23. Based on the boundary layer thickness information obtained in step S21, and combined with the classification results of the cluster centers in step S22, the boundary layer thickness is further adaptively updated; the membership degree is greater than the threshold. Data points sample points form a set The length is denoted as Boundary layer thickness Calculated by the following formula: in It is a cost function; and This represents the number of data points, with values ​​greater than and less than the cluster centers, respectively. It is the proportionality coefficient.

7. The method for optimizing drive anti-skid control based on vehicle-road cooperative road surface fusion estimation according to claim 6, characterized in that, The tire longitudinal force estimator and the longitudinal road surface estimator are designed as follows: In the formula, For tire longitudinal force estimation, To obtain the adhesion coefficient by substituting the road surface estimate and slip ratio into the tire model fitting formula, For the gain of the longitudinal force estimator, This is the numerical solution for the peak tire-road adhesion coefficient obtained by solving the tire model fitting formula based on the estimated longitudinal force and slip ratio. This is an estimate of the peak tire-road adhesion coefficient. For the road surface estimator gain, These are intermediate variables used in the process of constructing the estimator; Based on the estimation of tire lateral force and self-aligning moment, a lateral road surface estimation algorithm is designed: In the formula, For steering wheel angle, To improve the torque-to-displacement ratio of the motor to the kingpin, This is the rotation ratio of the steering wheel torque to the steering torque at the kingpin. Steering wheel torque, To assist the motor torque, A and B These are the fitting parameters for a second-order system. The main return torque. The estimated value of the normalizing moment is calculated based on the vertical load and the sideslip angle. For wheel mechanical trail, This is an estimate of the lateral acceleration. and For the estimator gain, This is the estimated value of the peak tire-road adhesion coefficient, i.e., the lateral force excitation. Based on the friction ellipse, the following assumptions can be established regarding the longitudinal and lateral adhesion limits of the tire: In the formula, For longitudinal adhesion limit, For lateral adhesion limit, Wheel slip angle The lateral adhesion limit of the contribution, Camber angle The lateral adhesion limit of the contribution, The peak adhesion coefficient, and The value is related to the vehicle status; Based on the above analysis, a dynamic fusion road surface estimator is designed as shown in the following equation, and the formulas in the original estimator are modified: In the formula, , for and The estimated value; The vision and dynamics fusion estimator is designed as follows: In the formula, This is a visual estimate of the road surface adhesion coefficient. This is the road surface estimation result from the previous moment. It is a dynamic estimation of gain that satisfies the local Lipchitz condition. It is a gain adjustment coefficient greater than 0. δ It is the boundary layer thickness. It is the road surface estimate after incorporating visual information.

8. The method for optimizing drive anti-skid control based on vehicle-road cooperative road surface fusion estimation according to claim 1, characterized in that, Step S3 specifically includes the following steps: S31. Using visual road surface recognition information as feedforward input, calculate the reference slip ratio and the actual slip ratio, while also incorporating wheel acceleration, thus expressing the controlled variable as the following formula: when At that time, the controlled variable is the wheel slip ratio. ,when At that time, the controlled variable is the equivalent wheel acceleration. ;when When the controlled variable is a linear combination of the two, substituting the wheel slip ratio and the reference slip ratio into the above equation, the output is... and ; S32. Use a conditional integral controller with feedforward: in , It is an integral term. for initial value, The coefficient before the integral term, The thickness of the synovial surface. This is the upper limit of the driving torque. The actual required torque after slip ratio control intervention; S33. Design the following control strategy based on the visual fusion results between the vehicle-mounted camera and the roadside camera. Predictive slip ratio controller output torque ,when At this time, the system performs slip ratio control, and the motor outputs a predicted control torque. Otherwise, then let Simultaneously, the following logic is executed: calculate the driver's required torque. Output torque of slip ratio controller If the ratio exceeds the set upper limit threshold... The system performs slip ratio control, and the motor outputs control torque; if this criterion is not triggered, the slip ratio controller enters a preload state, calculates and saves the slip ratio controller output torque. When the traditional slip criterion is met, the system immediately switches to preload control mode and quickly outputs the slip ratio controller output torque. This enables a low-latency response to the slip state; Traditional slip criteria are based on wheel angular acceleration With longitudinal acceleration The difference, and the preload torque Torque required by current drivers The ratios are used to construct angular acceleration and torque criteria, respectively; The angular acceleration flag is set in the initial state. ,when And duration At that time, the flag position At this point, the system will continuously monitor whether the conditions are met. And duration If satisfied, then the flag bit If the above judgment conditions are not triggered, the flag bit state remains unchanged; The torque flag is set in the initial state. ,when At that time, the flag position At this point, the system will continuously monitor whether the conditions are met. If satisfied, then the flag bit In all other cases, the flag state remains unchanged; the final torque output is determined by the combined state of the two flags, and the specific judgment logic is as follows: 。 9. The method for optimizing drive anti-skid control based on vehicle-road cooperative road surface fusion estimation according to claim 8, characterized in that, , The value is 1.

05. The value is 0.95, and S is 1. , Value is 30ms.

Images