Vehicle braking control method, electronic device and vehicle

Abstract

The present invention discloses a vehicle braking control method, electronic device and vehicle. The method comprises: obtaining the sideslip angle, actual yaw rate and actual longitudinal speed of the vehicle; calculating the reference yaw rate and reference longitudinal speed according to a pre-established two-degree-of-freedom vehicle model; determining the full state quantity of the braking system according to the sideslip angle, the actual yaw rate, the actual longitudinal speed, the reference yaw rate and the reference longitudinal speed; inputting the full state quantity into a pre-trained braking control model to obtain the braking torque output by the braking control model; and performing braking control on the vehicle according to the braking torque. The present invention uses the full state quantity of the braking system as the input of the EMB model by means of deep learning, optimizes the control according to the reward value of the learning network based on the S‑DDPG optimization strategy, realizes the ideal braking control performance index and reduces the actual system safety risk.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicles, and in particular relates to a vehicle braking control method, electronic equipment and a vehicle. Background Art

[0002] Unlike conventional hydraulic braking, the EMB (Electro-Mechanical Braking) braking system completely abandons the fluid energy-carrying medium and directly drives the actuator through the motor to generate the brake clamping force. In the current technical solution, the PID (Proportion Integral Differential) algorithm is usually applied to control the actual slip rate and the optimal slip rate based on the accurate vehicle speed estimation to achieve the control of the EMB braking system.

[0003] With the further development of autonomous driving of new energy vehicles, higher requirements are placed on the braking response speed and control accuracy of the braking system. Traditional parameter adjustment based on PID control can no longer meet the requirements for response speed and control accuracy. Summary of the invention

[0004] An object of the present invention is to provide a vehicle braking control method, an electronic device and a vehicle, which can solve the technical problems of slow response speed and low control accuracy of parameter adjustment based on PID control in the prior art.

[0005] According to a first aspect of the present invention, there is provided a vehicle braking control method, comprising:

[0006] Obtain the vehicle's sideslip angle, actual yaw rate, and actual longitudinal speed;

[0007] According to the pre-established two-degree-of-freedom vehicle model, a reference yaw rate and a reference longitudinal vehicle speed are calculated;

[0008] Determine a full state quantity of a braking system according to the sideslip angle, the actual yaw rate, the actual longitudinal vehicle speed, the reference yaw rate, and the reference longitudinal vehicle speed;

[0009] Inputting the full state quantity into a pre-trained braking control model to obtain a braking torque output by the braking control model, wherein the braking control model is a reinforcement learning model based on an S-DDPG algorithm and is obtained by training with a pre-defined reward and punishment function;

[0010] The vehicle is braked and controlled according to the braking torque.

[0011] Optionally, determining the full state quantity of the braking system according to the sideslip angle, the actual yaw rate, the actual longitudinal vehicle speed, the reference yaw rate, and the reference longitudinal vehicle speed includes:

[0012] Calculating a difference between the actual yaw rate and the reference yaw rate;

[0013] calculating a difference between the actual longitudinal vehicle speed and the reference longitudinal vehicle speed;

[0014] Determining the total state quantity;

[0015] ;

[0016] in, is the total state quantity, is the sideslip angle, is the actual yaw angular velocity, is the difference between the actual yaw rate and the reference yaw rate, is the actual longitudinal speed, It is the difference between the actual longitudinal speed and the reference longitudinal speed.

[0017] Optionally, the braking control model includes an action network and a value network, the input of the action network is the full state quantity, and the output of the action network is the braking torque;

[0018] The input of the value network is the full state quantity and the braking torque output by the action network, and the output of the value network is the Q value corresponding to the braking torque output by the action network, which is used to reflect the quality of the braking torque output by the action network.

[0019] Optionally, the reward and punishment function is:

[0020] ;

[0021] in, is the total penalty value, and They are a weighted quadratic reward and penalty function including vehicle body yaw angular acceleration, yaw angular velocity, and vehicle body acceleration, and a reward and penalty function combined with active safety;

[0022] Weighted quadratic reward and penalty function It is expressed as:

[0023] ;

[0024] in, is the vehicle speed, is the vehicle acceleration, is the yaw angular velocity, is the yaw angular acceleration;

[0025] Reward and Penalty Function Combined with Active Safety It is expressed as:

[0026] ;

[0027] ;

[0028] ;

[0029] ;

[0030] ;

[0031] ;

[0032] in, , , , are the influencing factors related to the sideslip angle, yaw rate, sideslip angle error limit and yaw rate error limit, C1, C2, C3 and C4 are the corresponding coefficients respectively. It is the representation of the vehicle body status. is the weight matrix representing the control performance.

[0033] Optionally, the action network includes an input layer, three hidden layers and an output layer, the input layer of the action network includes 8 nodes, the 8 nodes of the input layer of the action network correspond to the full state of the vehicle, each hidden layer of the action network includes 100 nodes, a full connection method is adopted, the activation function is a ReLU function, the output layer of the action network includes 4 nodes, and the activation function of the output layer of the action network is a Tanh function;

[0034] The value network includes an input layer, three hidden layers and an output layer. The input layer of the value network includes 12 nodes, and each hidden layer of the value network includes 100 nodes. A full connection method is adopted, and the activation function is the ReLU function. The output layer of the value network includes 1 node.

[0035] Optionally, before inputting the full state quantity into a pre-trained braking control model, the method further includes:

[0036] Initializing various weight parameters of the action network and the value network;

[0037] Define sample pools;

[0038] Fill the sample pool with samples generated by pseudo-labels;

[0039] Update the weight parameters through experience samples;

[0040] Obtain the full state quantity of the braking system at the current moment;

[0041] Obtain the action corresponding to the full state quantity of the braking system at the current moment through the action network;

[0042] Execute the action to obtain the full state quantity at the next moment and the observed value of the reward and penalty at the current moment, and update the full state quantity at the current moment according to the full state quantity at the next moment;

[0043] Take samples from the sample pool and calculate Q through the value network;

[0044] Calculate the loss function of the model;

[0045] Update the weight parameters of the value network through the gradient descent method;

[0046] Calculate the action network gradient and update the weight parameters of the action network;

[0047] If the preset constraints are met, the training is stopped and the trained braking control model is obtained;

[0048] If the preset constraints are not met, the steps of obtaining the action corresponding to the full state quantity at the current moment through the action network are repeated.

[0049] Optionally, the pseudo-label generation step includes:

[0050] Obtain source domain dataset and target dataset;

[0051] Determine the local density of each sample ;

[0052] ;

[0053] ;

[0054] in, Indicates that the distance is less than The number of samples, Depend on Elements identified;

[0055] The sample with the largest local density is taken as the cluster center, and the target data set is divided into several clusters according to the distance between the target sample and the cluster center and the clustering constraints to generate data subsets;

[0056] Train the source domain dataset and the target dataset to obtain the feature extractor , Source Classifier , Domain Classifier , target classifier ;

[0057] According to the source classifier And the loss function Generate pseudo labels ;

[0058] ;

[0059] ;

[0060] ;

[0061] ;

[0062] in, represents a pseudo-label sample set consisting of target samples with high density values, is the pseudo label of the target sample, is the source domain data, For the sample set The number of samples, For the sample set The number of samples, is the cross entropy loss, is the binary cross entropy loss, and are the characteristics of the source sample and the target sample respectively, To balance the parameters, is the source sample label, is a hyperparameter.

[0063] Optionally, the clustering constraint condition is:

[0064] ;

[0065] ;

[0066] in, Represents the target classifier The output of the i-th mini-batch gradient descent training on the softmax layer, express and similar, express and Different categories, express and The Euclidean distance between .

[0067] According to a second aspect of the present invention, an electronic device is provided, comprising a processor and a memory, wherein the memory stores programs or instructions that can be run on the processor, and when the programs or instructions are executed by the processor, the steps of a vehicle braking control method as described in the first aspect of the present invention are implemented.

[0068] According to a third aspect of the present invention, a vehicle is provided, comprising the electronic device according to the second aspect of the present invention.

[0069] The beneficial effects of the present invention are as follows: the present invention uses the full state quantity of the EMB brake system as the input of the EMB model by means of deep learning, optimizes the control according to the reward value of the learning network based on the S-DDPG optimization strategy, achieves ideal braking control performance indicators and reduces the actual system safety risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a schematic diagram of the EMB control strategy based on continuous control of slip rate error in the prior art.

[0071] Figure 2 It is a schematic diagram of an EMB simulation model based on continuous control of slip rate error in the prior art.

[0072] Figure 3 It is a flow chart of a vehicle braking control method in an embodiment of the present invention.

[0073] Figure 4 Schematic diagram of a two-degree-of-freedom model of a vehicle in an embodiment of the present invention.

[0074] Figure 5 Schematic diagram of a vehicle braking model in an embodiment of the present invention.

[0075] Figure 6 It is a schematic diagram of the convolution accelerator architecture based on FPGA in an embodiment of the present invention. DETAILED DESCRIPTION

[0076] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless otherwise specifically stated.

[0077] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0078] The techniques, methods and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the techniques, methods and devices should be considered as part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0079] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0080] In the specification and claims of the present invention, if the term "first" or "second" is involved, it may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "plurality" means two or more. In addition, "and / or" in the specification and claims means at least one of the connected objects, and the character " / " generally means that the related objects are in an "or" relationship.

[0081] Unlike conventional hydraulic brakes, the EMB brake system completely abandons the fluid energy medium and directly drives the actuator to generate the brake clamping force through the motor. In the current technical solution, the PID algorithm is usually applied to control the actual slip rate and the optimal slip rate based on the accurate vehicle speed estimation to realize the control of the EMB brake system. Its control framework and Simulink control model are as follows Figure 1 and Figure 2 shown.

[0082] With the further development of autonomous driving of new energy vehicles, higher requirements are placed on the braking response speed and control accuracy of the braking system. Traditional parameter adjustment based on PID control can no longer meet the requirements for response speed and control accuracy.

[0083] like Figure 3 As shown, this embodiment introduces a vehicle braking control method, including steps 1100-1500.

[0084] Step 1100: Obtain the sideslip angle, actual yaw rate, and actual longitudinal speed of the vehicle.

[0085] The sideslip angle, actual yaw rate, and actual longitudinal vehicle speed can be obtained through various sensors installed on the vehicle.

[0086] Step 1200: Calculate a reference yaw rate and a reference longitudinal vehicle speed according to a pre-established two-degree-of-freedom vehicle model.

[0087] The two-degree-of-freedom vehicle model is Figure 4 shown. Figure 4 Central Origin represents the center of mass of the vehicle, Indicates the lateral force along the Y-axis on the front wheel of the vehicle. Indicates the lateral force along the X-axis on the front wheel of the vehicle. Indicates the lateral force along the Y-axis acting on the rear wheels of the vehicle. represents the sideslip angle of the front wheel, Indicates the sideslip angle of the rear wheels. Represents the center of the front wheel and the center of mass of the vehicle The distance between Represents the center of the rear wheel and the center of mass of the vehicle The distance between.

[0088] .

[0089] .

[0090] in, Indicates the mass of the vehicle, is the yaw angular velocity, Indicates the steering angle. For vehicle speed Component along the X axis. Indicates the acceleration of the vehicle. represents the yaw angular acceleration.

[0091] Front wheel slip angle It is expressed as:

[0092] .

[0093] Rear wheel slip angle It is expressed as:

[0094] .

[0095] Step 1300: Determine the full state quantity of the braking system according to the sideslip angle, the actual yaw rate, the actual longitudinal vehicle speed, the reference yaw rate, and the reference longitudinal vehicle speed.

[0096] Step 1400: Input the full state quantity into a pre-trained braking control model to obtain a braking torque output by the braking control model, wherein the braking control model is a reinforcement learning model based on the S-DDPG algorithm, which is trained by a pre-defined reward and punishment function.

[0097] DDPG is a reinforcement learning algorithm that uses neural networks to approximate the optimal policy and stabilizes the learning process through techniques such as experience replay and target networks. Safe DDPG introduces safety constraints to ensure that the agent does not violate the preset safety boundaries during the decision-making process, thereby ensuring the safety of the learning process.

[0098] Safe Deep Deterministic Policy Gradient (Safe DDPG, S-DDPG) is a method that combines deep learning and reinforcement learning to solve continuous control problems while ensuring the safety of the learning process. This algorithm is based on the Deep Deterministic Policy Gradient (DDPG) algorithm and is extended to meet the needs of safe learning.

[0099] Step 1500: Perform braking control on the vehicle according to the braking torque.

[0100] The present invention uses the full state quantity of the EMB brake system as the input of the EMB model by means of deep learning, and based on the S-DDPG optimization strategy, optimizes the control according to the reward value of the learning network to achieve ideal braking control performance indicators and reduce the actual system safety risks.

[0101] In this embodiment, the step 1300 includes: calculating the difference between the actual yaw rate and the reference yaw rate; calculating the difference between the actual longitudinal vehicle speed and the reference longitudinal vehicle speed; and determining the full state quantity.

[0102] The full state of the brake system It can be expressed as:

[0103] .

[0104] in, is the total state quantity, is the sideslip angle, is the actual yaw angular velocity, is the difference between the actual yaw rate and the reference yaw rate, is the actual longitudinal speed, It is the difference between the actual longitudinal speed and the reference longitudinal speed.

[0105] In this embodiment, the braking control model includes an action network and a value network, the input of the action network is the full state quantity, and the output of the action network is the braking torque; the input of the value network is the full state quantity and the braking torque output by the action network, and the output of the value network is the Q value corresponding to the braking torque output by the action network, which is used to reflect the quality of the braking torque output by the action network.

[0106] like Figure 5 As shown, the input of the action network is the full state quantity , the action output by the action network Indicates the braking torque.

[0107] .

[0108] in represents the braking torque of the left front wheel, Indicates the braking torque of the right front wheel, represents the braking torque of the left rear wheel, Indicates the braking torque of the right rear wheel.

[0109] The Q value output by the value network represents the action value, which is used to evaluate the quality of taking an action under a specific state. In this embodiment, the Q value output by the value network can be expressed as:

[0110] .

[0111] in, Indicates in status and actions The Q value output by the next value network is, is the total state quantity at time t, The output of the action network and the total state The corresponding action, Represents the observed value of reward and penalty at time t+1.

[0112] In this embodiment, the reward and punishment function is:

[0113] .

[0114] in, is the total penalty value, and They are respectively a weighted quadratic reward and punishment function including vehicle body yaw angular acceleration, yaw angular velocity, and vehicle body acceleration, and a reward and punishment function combined with active safety.

[0115] Weighted quadratic reward and penalty function It is expressed as:

[0116] ;

[0117] in, is the vehicle speed, is the vehicle acceleration, is the yaw angular velocity, is the yaw angular acceleration.

[0118] Reward and Penalty Function Combined with Active Safety It is expressed as:

[0119] ;

[0120] ;

[0121] ;

[0122] ;

[0123] ;

[0124] ;

[0125] in, , , , are the influencing factors related to the sideslip angle, yaw rate, sideslip angle error limit and yaw rate error limit, C1, C2, C3 and C4 are the corresponding coefficients respectively. It is the representation of the vehicle body status. is the weight matrix representing the control performance.

[0126] In the above reward and punishment function, the sideslip angle, yaw angular velocity and yaw angular velocity error are taken into account, and can be adjusted according to different states. On the basis of ensuring the accuracy of network calculation, driving safety is also taken into account.

[0127] In this embodiment, the action network includes an input layer, three hidden layers and an output layer. The input layer of the action network includes 8 nodes, and the 8 nodes of the input layer of the action network correspond to the full state of the vehicle. Each hidden layer of the action network includes 100 nodes, and a full connection method is adopted. The activation function is the ReLU function. The output layer of the action network includes 4 nodes, and the activation function of the output layer of the action network is the Tanh function.

[0128] The value network includes an input layer, three hidden layers and an output layer. The input layer of the value network includes 12 nodes, and each hidden layer of the value network includes 100 nodes. A full connection method is adopted, and the activation function is the ReLU function. The output layer of the value network includes 1 node.

[0129] like Figure 5 As shown, the action network includes an input layer, a hidden layer, and an output layer. The 8 nodes of the input layer of the action network correspond to the 8 parameters in the full state of the braking system. The 4 nodes of the output layer of the action network correspond to the braking torque of the 4 wheels of the vehicle.

[0130] The value network includes an input layer, three hidden layers and an output layer. The 12 nodes in the input layer of the value network correspond to the eight parameters in the full state of the braking system and the braking torque of the four wheels output by the action network.

[0131] In this embodiment, before the full state quantity is input into the pre-trained braking control model, the method further includes steps S1-S10. Steps S1-S10 represent the training process of the braking control model.

[0132] Step S1: Initialize the weight parameters of the action network and the value network.

[0133] The weight parameters of the action network and the value network can be randomly initialized, or the initialization values ​​of the weight parameters can be set based on experience.

[0134] Step S2: Define a sample pool, fill the sample pool with samples generated by pseudo labels, and update the weight parameters with empirical samples.

[0135] Pseudo-labeling is to predict unlabeled data through the model and use the prediction results as pseudo-labels for the training process. The core idea of ​​this method is to use the model itself to classify unlabeled data and use the classification results as additional training data, thereby expanding the scale of the training set and improving the generalization ability of the model.

[0136] Updating the weight parameters through experience samples can make the weight parameters more accurate and improve the efficiency of model training.

[0137] Step S3: Obtain the current state of the braking system.

[0138] Step S4: Obtain the action corresponding to the full state quantity of the braking system at the current moment through the action network.

[0139] Assuming the current time is t, the total state quantity at the current time is recorded as , the corresponding action output by the action network is .

[0140] Step S5: Execute the action to obtain the total state quantity at the next moment and the observed value of the reward and penalty at the current moment, and update the total state quantity at the current moment according to the total state quantity at the next moment.

[0141] The next moment is t+1, and the action is executed Then get the full state quantity at the next moment The observed value of reward and punishment can be calculated according to the reward and punishment function. Then the state is updated to .

[0142] Step S6: Take samples from the sample pool and calculate Q through the value network.

[0143] Step S7: Calculate the loss function of the model.

[0144] Step S8: Update the weight parameters of the value network through the gradient descent method.

[0145] Step S9: Calculate the action network gradient and update the weight parameters of the action network.

[0146] Step S10: If the preset constraints are met, stop the training and obtain the trained braking control model; if the preset constraints are not met, repeat the step of obtaining the action corresponding to the full state quantity at the current moment through the action network.

[0147] The loss function of the model is used to measure the inconsistency between the model's predicted value and the true value. The smaller the loss function, the better the robustness of the model. The loss function can be used as a constraint. When the calculated loss function meets the pre-set standard, the training ends. If not, continue to iterate and repeat steps S4-S9.

[0148] In this embodiment, the pseudo-label generation step includes:

[0149] Obtain source domain dataset and target dataset;

[0150] Determine the local density of each sample ;

[0151] ;

[0152] ;

[0153] in, Indicates that the distance is less than The number of samples, Depend on Elements identified;

[0154] The sample with the largest local density is taken as the cluster center, and the target data set is divided into several clusters according to the distance between the target sample and the cluster center and the clustering constraints to generate data subsets;

[0155] Train the source domain dataset and the target dataset to obtain the feature extractor , Source Classifier , Domain Classifier , target classifier ;

[0156] According to the source classifier And the loss function Generate pseudo labels ;

[0157] ;

[0158] ;

[0159] ;

[0160] ;

[0161] in, represents a pseudo-label sample set consisting of target samples with high density values, is the pseudo label of the target sample, is the source domain data, For the sample set The number of samples, For the sample set The number of samples, is the cross entropy loss, is the binary cross entropy loss, and are the characteristics of the source sample and the target sample respectively, To balance the parameters, is the source sample label, is a hyperparameter.

[0162] In this embodiment, the clustering constraint condition is:

[0163] ;

[0164] ;

[0165] in, Represents the target classifier The output of the i-th mini-batch gradient descent training on the softmax layer, express and similar, express and Different categories, express and The Euclidean distance between .

[0166] like Figure 6 As shown, this embodiment introduces a vehicle braking control device, including an FPGA (Field Programmable Gate Array) chip, a torque distribution control strategy algorithm based on S-DDPG is deployed on the FPGA chip, and network parameters are stored in an off-chip DDR memory. During calculation, the required parameters are loaded into the FPGA chip, and the DDR read and write times and the overall system utilization are optimized according to different parallel configurations, and a hardware accelerator is designed.

[0167] This embodiment introduces an electronic device, including a processor and a memory, wherein the memory stores programs or instructions that can be run on the processor, and when the program or instructions are executed by the processor, the steps of a vehicle braking control method as described in any embodiment of the present invention are implemented.

[0168] This embodiment introduces a vehicle, including the electronic device described in the above embodiment of the present invention.

[0169] Although some specific embodiments of the present invention have been described in detail by way of example, it will be appreciated by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It will be appreciated by those skilled in the art that the above embodiments may be modified without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

[0170] Those of ordinary skill in the art will appreciate that the modules and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0171] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and equipment can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0172] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0173] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of the present invention.

[0174] In addition, each functional module in the embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0175] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the energy-saving signal sending / receiving method of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

[0176] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.

[0177] It should be understood that the size of the sequence number of each step in the content of the invention and the embodiments of the present invention does not absolutely mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention. For the purpose of example and description, the foregoing description of the implementation of the present disclosure has been given. The foregoing description is not exhaustive and is not intended to limit the present disclosure to the exact form disclosed. Various deformations and modifications may exist according to the above teachings, or various deformations and modifications may be obtained from the practice of the present disclosure. These embodiments are selected and described to illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can use the present disclosure in various embodiments and various modifications suitable for the specific purpose conceived.

Claims

1. A vehicle braking control method, characterized in that: include: Obtain the vehicle's sideslip angle, actual yaw rate, and actual longitudinal speed; According to the pre-established two-degree-of-freedom vehicle model, a reference yaw rate and a reference longitudinal vehicle speed are calculated; Determine a full state quantity of a braking system according to the sideslip angle, the actual yaw rate, the actual longitudinal vehicle speed, the reference yaw rate, and the reference longitudinal vehicle speed; Inputting the full state quantity into a pre-trained braking control model to obtain a braking torque output by the braking control model, wherein the braking control model is a reinforcement learning model based on an S-DDPG algorithm and is obtained by training with a pre-defined reward and punishment function; performing braking control on the vehicle according to the braking torque; The reward and punishment function is: ； in, is the total penalty value, and They are a weighted quadratic reward and penalty function including vehicle body yaw angular acceleration, yaw angular velocity, and vehicle body acceleration, and a reward and penalty function combined with active safety; Weighted quadratic reward and penalty function It is expressed as: ； in, is the vehicle speed, is the vehicle acceleration, is the yaw angular velocity, is the yaw angular acceleration; Reward and Penalty Function Combined with Active Safety It is expressed as: ； ； ； ； ； ； in, , , , are the influencing factors related to the sideslip angle, yaw rate, sideslip angle error limit and yaw rate error limit, C1, C2, C3 and C4 are the corresponding coefficients respectively. It is the representation of the vehicle body status. is the weight matrix representing the control performance.

2. The method according to claim 1, characterized in that The method of determining the full state quantity of the braking system according to the sideslip angle, the actual yaw rate, the actual longitudinal vehicle speed, the reference yaw rate, and the reference longitudinal vehicle speed includes: Calculating a difference between the actual yaw rate and the reference yaw rate; calculating a difference between the actual longitudinal vehicle speed and the reference longitudinal vehicle speed; Determining the total state quantity; ； in, is the total state quantity, is the sideslip angle, is the actual yaw angular velocity, is the difference between the actual yaw rate and the reference yaw rate, is the actual longitudinal speed, It is the difference between the actual longitudinal speed and the reference longitudinal speed.

3. The method according to claim 2, characterized in that The braking control model includes an action network and a value network, the input of the action network is the full state quantity, and the output of the action network is the braking torque; The input of the value network is the full state quantity and the braking torque output by the action network, and the output of the value network is the Q value corresponding to the braking torque output by the action network, which is used to reflect the quality of the braking torque output by the action network.

4. The method according to claim 3, characterized in that The action network includes an input layer, three hidden layers and an output layer. The input layer of the action network includes 8 nodes. The 8 nodes of the input layer of the action network correspond to the full state of the vehicle. Each hidden layer of the action network includes 100 nodes. A full connection method is adopted. The activation function is a ReLU function. The output layer of the action network includes 4 nodes. The activation function of the output layer of the action network is a Tanh function. The value network includes an input layer, three hidden layers and an output layer. The input layer of the value network includes 12 nodes, and each hidden layer of the value network includes 100 nodes. A full connection method is adopted, and the activation function is the ReLU function. The output layer of the value network includes 1 node.

5. The method according to claim 3, characterized in that: Before inputting the full state quantity into the pre-trained braking control model, the method further includes: Initializing various weight parameters of the action network and the value network; Define sample pools; Fill the sample pool with samples generated by pseudo-labels; Update the weight parameters through experience samples; Obtain the full state quantity of the braking system at the current moment; Obtain the action corresponding to the full state quantity of the braking system at the current moment through the action network; Execute the action to obtain the full state quantity at the next moment and the observed value of the reward and penalty at the current moment, and update the full state quantity at the current moment according to the full state quantity at the next moment; Take samples from the sample pool and calculate Q through the value network; Calculate the loss function of the model; Update the weight parameters of the value network through the gradient descent method; Calculate the action network gradient and update the weight parameters of the action network; If the preset constraints are met, the training is stopped and the trained braking control model is obtained; If the preset constraints are not met, the steps of obtaining the action corresponding to the full state quantity at the current moment through the action network are repeated.

6. The method according to claim 5, characterized in that The pseudo-label generation step comprises: Obtain source domain dataset and target dataset; Determine the local density of each sample ; ； ； in, Indicates that the distance is less than The number of samples, Depend on Elements identified; The sample with the largest local density is taken as the cluster center, and the target data set is divided into several clusters according to the distance between the target sample and the cluster center and the clustering constraints to generate data subsets; Train the source domain dataset and the target dataset to obtain the feature extractor , Source Classifier , Domain Classifier , target classifier ; According to the source classifier And the loss function Generate pseudo labels ; ； ； ； ； in, represents a pseudo-label sample set consisting of target samples with high density values, is the pseudo label of the target sample, is the source domain data, For the sample set The number of samples, For the sample set The number of samples, is the cross entropy loss, is the binary cross entropy loss, and are the characteristics of the source sample and the target sample respectively, To balance the parameters, is the source sample label, is a hyperparameter.

7. The method according to claim 6, characterized in that The clustering constraints are: ； ； in, Represents the target classifier The output of the i-th mini-batch gradient descent training on the softmax layer, express and similar, express and Different categories, express and The Euclidean distance between .

8. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the steps of a vehicle braking control method as described in any one of claims 1 to 7 are implemented.

9. A vehicle, characterized in that: The electronic device comprising the electronic device described in claim 8.

Images